├── .idea ├── .gitignore ├── Yolov5_DeepSort_Pytorch.iml ├── inspectionProfiles │ ├── Project_Default.xml │ └── profiles_settings.xml ├── misc.xml ├── modules.xml ├── other.xml ├── vcs.xml └── workspace.xml ├── LICENSE ├── README.md ├── Town.gif ├── __pycache__ └── track.cpython-38.pyc ├── app.py ├── camera1.py ├── deep_sort ├── .gitignore ├── LICENSE ├── README.md ├── configs │ ├── deep_sort.yaml │ ├── yolov3.yaml │ └── yolov3_tiny.yaml ├── deep_sort │ ├── README.md │ ├── __init__.py │ ├── deep │ │ ├── __init__.py │ │ ├── checkpoint │ │ │ └── .gitkeep │ │ ├── evaluate.py │ │ ├── feature_extractor.py │ │ ├── model.py │ │ ├── original_model.py │ │ ├── test.py │ │ ├── train.jpg │ │ └── train.py │ ├── deep_sort.py │ └── sort │ │ ├── __init__.py │ │ ├── detection.py │ │ ├── iou_matching.py │ │ ├── kalman_filter.py │ │ ├── linear_assignment.py │ │ ├── nn_matching.py │ │ ├── preprocessing.py │ │ ├── track.py │ │ └── tracker.py ├── demo │ ├── 1.jpg │ ├── 2.jpg │ └── demo.gif ├── detector │ ├── YOLOv3 │ │ ├── README.md │ │ ├── __init__.py │ │ ├── cfg.py │ │ ├── cfg │ │ │ ├── coco.data │ │ │ ├── coco.names │ │ │ ├── darknet19_448.cfg │ │ │ ├── tiny-yolo-voc.cfg │ │ │ ├── tiny-yolo.cfg │ │ │ ├── voc.data │ │ │ ├── voc.names │ │ │ ├── voc_gaotie.data │ │ │ ├── yolo-voc.cfg │ │ │ ├── yolo.cfg │ │ │ ├── yolo_v3.cfg │ │ │ └── yolov3-tiny.cfg │ │ ├── darknet.py │ │ ├── demo │ │ │ ├── 004545.jpg │ │ │ └── results │ │ │ │ └── 004545.jpg │ │ ├── detect.py │ │ ├── detector.py │ │ ├── nms │ │ │ ├── __init__.py │ │ │ ├── build.sh │ │ │ ├── ext │ │ │ │ ├── __init__.py │ │ │ │ ├── build.py │ │ │ │ ├── cpu │ │ │ │ │ ├── nms_cpu.cpp │ │ │ │ │ └── vision.h │ │ │ │ ├── cuda │ │ │ │ │ ├── nms.cu │ │ │ │ │ └── vision.h │ │ │ │ ├── nms.h │ │ │ │ └── vision.cpp │ │ │ ├── nms.py │ │ │ └── python_nms.py │ │ ├── region_layer.py │ │ ├── weight │ │ │ └── .gitkeep │ │ ├── yolo_layer.py │ │ └── yolo_utils.py │ └── __init__.py ├── ped_det_server.py ├── scripts │ ├── yolov3_deepsort.sh │ └── yolov3_tiny_deepsort.sh ├── utils │ ├── __init__.py │ ├── asserts.py │ ├── draw.py │ ├── evaluation.py │ ├── io.py │ ├── json_logger.py │ ├── log.py │ ├── parser.py │ └── tools.py ├── webserver │ ├── .env │ ├── __init__.py │ ├── config │ │ └── config.py │ ├── images │ │ ├── Thumbs.db │ │ ├── arc.png │ │ └── request.png │ ├── readme.md │ ├── rtsp_threaded_tracker.py │ ├── rtsp_webserver.py │ ├── server_cfg.py │ └── templates │ │ └── index.html ├── yolov3_deepsort.py └── yolov3_deepsort_eval.py ├── requirements.txt ├── templates └── index.html ├── track.py ├── track_demo.py └── yolov5 ├── .dockerignore ├── .gitattributes ├── .github ├── ISSUE_TEMPLATE │ ├── --bug-report.md │ ├── --feature-request.md │ └── -question.md └── workflows │ ├── ci-testing.yml │ ├── greetings.yml │ ├── rebase.yml │ └── stale.yml ├── .gitignore ├── Dockerfile ├── LICENSE ├── README.md ├── detect.py ├── hubconf.py ├── models ├── __init__.py ├── common.py ├── experimental.py ├── export.py ├── hub │ ├── yolov3-spp.yaml │ ├── yolov5-fpn.yaml │ └── yolov5-panet.yaml ├── yolo.py ├── yolov5l.yaml ├── yolov5m.yaml ├── yolov5s.yaml └── yolov5x.yaml ├── requirements.txt ├── test.py ├── train.py ├── tutorial.ipynb ├── utils ├── __init__.py ├── activations.py ├── datasets.py ├── evolve.sh ├── general.py ├── google_utils.py └── torch_utils.py └── weights └── download_weights.sh /.idea/.gitignore: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/.idea/.gitignore -------------------------------------------------------------------------------- /.idea/Yolov5_DeepSort_Pytorch.iml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 12 | 13 | 19 | 20 | 21 | 23 | -------------------------------------------------------------------------------- /.idea/inspectionProfiles/Project_Default.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 33 | -------------------------------------------------------------------------------- /.idea/inspectionProfiles/profiles_settings.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 7 | -------------------------------------------------------------------------------- /.idea/misc.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | -------------------------------------------------------------------------------- /.idea/modules.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | -------------------------------------------------------------------------------- /.idea/other.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 6 | -------------------------------------------------------------------------------- /.idea/vcs.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # PyQt5_YoLoV5_DeepSort 2 | 3 | 4 | This is a PyQt5 GUI program, which is based on YoloV5 and DeepSort to track person, and give a warning when someone comes to the "dangerous area". 5 | ======= 6 | This is a PyQt5 GUI program, which is based on YoloV5 and DeepSort to track person, and give a warning when someone comes to the "dangerous area". 7 | -------------------------------------------------------------------------------- /Town.gif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/Town.gif -------------------------------------------------------------------------------- /__pycache__/track.cpython-38.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/__pycache__/track.cpython-38.pyc -------------------------------------------------------------------------------- /camera1.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python 2 | # -*- coding: UTF-8 -*- 3 | import sys 4 | import cv2 5 | import torch 6 | from PyQt5 import QtCore, QtGui, QtWidgets 7 | from PyQt5.QtWidgets import * 8 | from track import detect 9 | 10 | 11 | class Ui_MainWindow(QtWidgets.QWidget): 12 | def __init__(self, parent=None): 13 | super(Ui_MainWindow, self).__init__(parent) 14 | self.CAM_NUM = 1 15 | self.timer_camera = QtCore.QTimer() 16 | self.s_rtsp = "rtsp://iscas:opqwer12@192.168.100.176:554/Streaming/Channels/101" 17 | self.cap = cv2.VideoCapture() 18 | 19 | self.set_ui() 20 | self.slot_init() 21 | self.__flag_work = 0 22 | self.x = 0 23 | self.count = 0 24 | 25 | def set_ui(self): 26 | 27 | self.__layout_main = QtWidgets.QHBoxLayout() 28 | self.__layout_fun_button = QtWidgets.QVBoxLayout() 29 | self.__layout_data_show = QtWidgets.QVBoxLayout() 30 | 31 | self.button_open_camera = QtWidgets.QPushButton(u'打开相机') 32 | 33 | self.button_detect = QtWidgets.QPushButton(u'开始检测') 34 | 35 | self.button_close = QtWidgets.QPushButton(u'退出') 36 | 37 | # Button 的颜色修改 38 | button_color = [self.button_open_camera, self.button_detect, self.button_close] 39 | for i in range(len(button_color)): 40 | button_color[i].setStyleSheet("QPushButton{color:black}" 41 | "QPushButton:hover{color:red}" 42 | "QPushButton{background-color:rgb(78,255,255)}" 43 | "QPushButton{border:2px}" 44 | "QPushButton{border-radius:10px}" 45 | "QPushButton{padding:2px 4px}") 46 | 47 | self.button_open_camera.setMinimumHeight(50) 48 | self.button_detect.setMinimumHeight(50) 49 | self.button_close.setMinimumHeight(50) 50 | 51 | # move()方法移动窗口在屏幕上的位置到x = 300，y = 300坐标。 52 | self.move(500, 500) 53 | 54 | # 信息显示 55 | self.label_show_camera = QtWidgets.QLabel() 56 | self.label_move = QtWidgets.QLabel() 57 | self.label_move.setFixedSize(100, 100) 58 | 59 | self.label_show_camera.setFixedSize(700, 600) 60 | self.label_show_camera.setAutoFillBackground(False) 61 | 62 | self.__layout_fun_button.addWidget(self.button_open_camera) 63 | self.__layout_fun_button.addWidget(self.button_detect) 64 | self.__layout_fun_button.addWidget(self.button_close) 65 | self.__layout_fun_button.addWidget(self.label_move) 66 | 67 | self.__layout_main.addLayout(self.__layout_fun_button) 68 | self.__layout_main.addWidget(self.label_show_camera) 69 | 70 | self.setLayout(self.__layout_main) 71 | self.label_move.raise_() 72 | self.setWindowTitle(u'摄像头') 73 | 74 | # 设置背景图片 75 | # palette1 = QPalette() 76 | # palette1.setBrush(self.backgroundRole(), QBrush(QPixmap('1.png'))) 77 | # self.setPalette(palette1) 78 | 79 | def slot_init(self): 80 | self.button_open_camera.clicked.connect(self.button_open_camera_click) 81 | self.timer_camera.timeout.connect(self.show_camera) 82 | self.button_close.clicked.connect(self.close) 83 | # self.button_detect.clicked.connect(self.button_detect_click) 84 | 85 | def button_detect_click(self): 86 | pass 87 | 88 | def button_open_camera_click(self): 89 | if not self.timer_camera.isActive(): 90 | flag = self.cap.open(self.s_rtsp) 91 | # flag = self.cap.isOpened() 92 | if not flag: 93 | msg = QtWidgets.QMessageBox.warning(self, u"Warning", u"请检测相机与电脑是否连接正确", 94 | buttons=QtWidgets.QMessageBox.Ok, 95 | defaultButton=QtWidgets.QMessageBox.Ok) 96 | # if msg==QtGui.QMessageBox.Cancel: 97 | # pass 98 | else: 99 | self.timer_camera.start(30) 100 | 101 | self.button_open_camera.setText(u'关闭相机') 102 | else: 103 | self.timer_camera.stop() 104 | self.cap.release() 105 | self.label_show_camera.clear() 106 | self.button_open_camera.setText(u'打开相机') 107 | 108 | def show_camera(self): 109 | flag, image = self.cap.read() 110 | 111 | show = cv2.resize(image, (640, 480)) 112 | show = cv2.cvtColor(show, cv2.COLOR_BGR2RGB) 113 | 114 | showImage = QtGui.QImage(show.data, show.shape[1], show.shape[0], QtGui.QImage.Format_RGB888) 115 | self.label_show_camera.setPixmap(QtGui.QPixmap.fromImage(showImage)) 116 | 117 | def closeEvent(self, event): 118 | ok = QtWidgets.QPushButton() 119 | cacel = QtWidgets.QPushButton() 120 | 121 | msg = QtWidgets.QMessageBox(QtWidgets.QMessageBox.Warning, u"关闭", u"确定退出？") 122 | 123 | msg.addButton(ok, QtWidgets.QMessageBox.ActionRole) 124 | msg.addButton(cacel, QtWidgets.QMessageBox.RejectRole) 125 | ok.setText(u'确定') 126 | cacel.setText(u'取消') 127 | # msg.setDetailedText('sdfsdff') 128 | if msg.exec_() == QtWidgets.QMessageBox.RejectRole: 129 | event.ignore() 130 | else: 131 | # self.socket_client.send_command(self.socket_client.current_user_command) 132 | if self.cap.isOpened(): 133 | self.cap.release() 134 | if self.timer_camera.isActive(): 135 | self.timer_camera.stop() 136 | event.accept() 137 | 138 | 139 | if __name__ == "__main__": 140 | App = QApplication(sys.argv) 141 | ex = Ui_MainWindow() 142 | ex.show() 143 | sys.exit(App.exec_()) 144 | -------------------------------------------------------------------------------- /deep_sort/.gitignore: -------------------------------------------------------------------------------- 1 | # Folders 2 | __pycache__/ 3 | build/ 4 | *.egg-info 5 | 6 | 7 | # Files 8 | *.weights 9 | *.t7 10 | *.mp4 11 | *.avi 12 | *.so 13 | *.txt 14 | -------------------------------------------------------------------------------- /deep_sort/LICENSE: -------------------------------------------------------------------------------- 1 | MIT License 2 | 3 | Copyright (c) 2020 Ziqiang 4 | 5 | Permission is hereby granted, free of charge, to any person obtaining a copy 6 | of this software and associated documentation files (the "Software"), to deal 7 | in the Software without restriction, including without limitation the rights 8 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 9 | copies of the Software, and to permit persons to whom the Software is 10 | furnished to do so, subject to the following conditions: 11 | 12 | The above copyright notice and this permission notice shall be included in all 13 | copies or substantial portions of the Software. 14 | 15 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 18 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 20 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 21 | SOFTWARE. -------------------------------------------------------------------------------- /deep_sort/README.md: -------------------------------------------------------------------------------- 1 | # Deep Sort with PyTorch 2 | 3 | ![](demo/demo.gif) 4 | 5 | ## Update(1-1-2020) 6 | Changes 7 | - fix bugs 8 | - refactor code 9 | - accerate detection by adding nms on gpu 10 | 11 | ## Latest Update(07-22) 12 | Changes 13 | - bug fix (Thanks @JieChen91 and @yingsen1 for bug reporting). 14 | - using batch for feature extracting for each frame, which lead to a small speed up. 15 | - code improvement. 16 | 17 | Futher improvement direction 18 | - Train detector on specific dataset rather than the official one. 19 | - Retrain REID model on pedestrain dataset for better performance. 20 | - Replace YOLOv3 detector with advanced ones. 21 | 22 | **Any contributions to this repository is welcome!** 23 | 24 | 25 | ## Introduction 26 | This is an implement of MOT tracking algorithm deep sort. Deep sort is basicly the same with sort but added a CNN model to extract features in image of human part bounded by a detector. This CNN model is indeed a RE-ID model and the detector used in [PAPER](https://arxiv.org/abs/1703.07402) is FasterRCNN , and the original source code is [HERE](https://github.com/nwojke/deep_sort). 27 | However in original code, the CNN model is implemented with tensorflow, which I'm not familier with. SO I re-implemented the CNN feature extraction model with PyTorch, and changed the CNN model a little bit. Also, I use **YOLOv3** to generate bboxes instead of FasterRCNN. 28 | 29 | ## Dependencies 30 | - python 3 (python2 not sure) 31 | - numpy 32 | - scipy 33 | - opencv-python 34 | - sklearn 35 | - torch >= 0.4 36 | - torchvision >= 0.1 37 | - pillow 38 | - vizer 39 | - edict 40 | 41 | ## Quick Start 42 | 0. Check all dependencies installed 43 | ```bash 44 | pip install -r requirements.txt 45 | ``` 46 | for user in china, you can specify pypi source to accelerate install like: 47 | ```bash 48 | pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple 49 | ``` 50 | 51 | 1. Clone this repository 52 | ``` 53 | git clone git@github.com:ZQPei/deep_sort_pytorch.git 54 | ``` 55 | 56 | 2. Download YOLOv3 parameters 57 | ``` 58 | cd detector/YOLOv3/weight/ 59 | wget https://pjreddie.com/media/files/yolov3.weights 60 | wget https://pjreddie.com/media/files/yolov3-tiny.weights 61 | cd ../../../ 62 | ``` 63 | 64 | 3. Download deepsort parameters ckpt.t7 65 | ``` 66 | cd deep_sort/deep/checkpoint 67 | # download ckpt.t7 from 68 | https://drive.google.com/drive/folders/1xhG0kRH1EX5B9_Iz8gQJb7UNnn_riXi6 to this folder 69 | cd ../../../ 70 | ``` 71 | 72 | 4. Compile nms module 73 | ```bash 74 | cd detector/YOLOv3/nms 75 | sh build.sh 76 | cd ../../.. 77 | ``` 78 | 79 | Notice: 80 | If compiling failed, the simplist way is to **Upgrade your pytorch >= 1.1 and torchvision >= 0.3" and you can avoid the troublesome compiling problems which are most likely caused by either `gcc version too low` or `libraries missing`. 81 | 82 | 5. Run demo 83 | ``` 84 | usage: python yolov3_deepsort.py VIDEO_PATH 85 | [--help] 86 | [--frame_interval FRAME_INTERVAL] 87 | [--config_detection CONFIG_DETECTION] 88 | [--config_deepsort CONFIG_DEEPSORT] 89 | [--display] 90 | [--display_width DISPLAY_WIDTH] 91 | [--display_height DISPLAY_HEIGHT] 92 | [--save_path SAVE_PATH] 93 | [--cpu] 94 | 95 | # yolov3 + deepsort 96 | python yolov3_deepsort.py [VIDEO_PATH] 97 | 98 | # yolov3_tiny + deepsort 99 | python yolov3_deepsort.py [VIDEO_PATH] --config_detection ./configs/yolov3_tiny.yaml 100 | 101 | # yolov3 + deepsort on webcam 102 | python3 yolov3_deepsort.py /dev/video0 --camera 0 103 | 104 | # yolov3_tiny + deepsort on webcam 105 | python3 yolov3_deepsort.py /dev/video0 --config_detection ./configs/yolov3_tiny.yaml --camera 0 106 | ``` 107 | Use `--display` to enable display. 108 | Results will be saved to `./output/results.avi` and `./output/results.txt`. 109 | 110 | All files above can also be accessed from BaiduDisk! 111 | linker：[BaiduDisk](https://pan.baidu.com/s/1YJ1iPpdFTlUyLFoonYvozg) 112 | passwd：fbuw 113 | 114 | ## Training the RE-ID model 115 | The original model used in paper is in original_model.py, and its parameter here [original_ckpt.t7](https://drive.google.com/drive/folders/1xhG0kRH1EX5B9_Iz8gQJb7UNnn_riXi6). 116 | 117 | To train the model, first you need download [Market1501](http://www.liangzheng.com.cn/Project/project_reid.html) dataset or [Mars](http://www.liangzheng.com.cn/Project/project_mars.html) dataset. 118 | 119 | Then you can try [train.py](deep_sort/deep/train.py) to train your own parameter and evaluate it using [test.py](deep_sort/deep/test.py) and [evaluate.py](deep_sort/deep/evalute.py). 120 | ![train.jpg](deep_sort/deep/train.jpg) 121 | 122 | ## Demo videos and images 123 | [demo.avi](https://drive.google.com/drive/folders/1xhG0kRH1EX5B9_Iz8gQJb7UNnn_riXi6) 124 | [demo2.avi](https://drive.google.com/drive/folders/1xhG0kRH1EX5B9_Iz8gQJb7UNnn_riXi6) 125 | 126 | ![1.jpg](demo/1.jpg) 127 | ![2.jpg](demo/2.jpg) 128 | 129 | 130 | ## References 131 | - paper: [Simple Online and Realtime Tracking with a Deep Association Metric](https://arxiv.org/abs/1703.07402) 132 | 133 | - code: [nwojke/deep_sort](https://github.com/nwojke/deep_sort) 134 | 135 | - paper: [YOLOv3](https://pjreddie.com/media/files/papers/YOLOv3.pdf) 136 | 137 | - code: [Joseph Redmon/yolov3](https://pjreddie.com/darknet/yolo/) 138 | -------------------------------------------------------------------------------- /deep_sort/configs/deep_sort.yaml: -------------------------------------------------------------------------------- 1 | DEEPSORT: 2 | REID_CKPT: "deep_sort/deep_sort/deep/checkpoint/ckpt.t7" 3 | MAX_DIST: 0.2 4 | MIN_CONFIDENCE: 0.3 5 | NMS_MAX_OVERLAP: 0.5 6 | MAX_IOU_DISTANCE: 0.7 7 | MAX_AGE: 70 8 | N_INIT: 3 9 | NN_BUDGET: 100 10 | 11 | -------------------------------------------------------------------------------- /deep_sort/configs/yolov3.yaml: -------------------------------------------------------------------------------- 1 | YOLOV3: 2 | CFG: "./detector/YOLOv3/cfg/yolo_v3.cfg" 3 | WEIGHT: "./detector/YOLOv3/weight/yolov3.weights" 4 | CLASS_NAMES: "./detector/YOLOv3/cfg/coco.names" 5 | 6 | SCORE_THRESH: 0.5 7 | NMS_THRESH: 0.4 8 | -------------------------------------------------------------------------------- /deep_sort/configs/yolov3_tiny.yaml: -------------------------------------------------------------------------------- 1 | YOLOV3: 2 | CFG: "./detector/YOLOv3/cfg/yolov3-tiny.cfg" 3 | WEIGHT: "./detector/YOLOv3/weight/yolov3-tiny.weights" 4 | CLASS_NAMES: "./detector/YOLOv3/cfg/coco.names" 5 | 6 | SCORE_THRESH: 0.5 7 | NMS_THRESH: 0.4 -------------------------------------------------------------------------------- /deep_sort/deep_sort/README.md: -------------------------------------------------------------------------------- 1 | # Deep Sort 2 | 3 | This is the implemention of deep sort with pytorch. -------------------------------------------------------------------------------- /deep_sort/deep_sort/__init__.py: -------------------------------------------------------------------------------- 1 | from .deep_sort import DeepSort 2 | 3 | 4 | __all__ = ['DeepSort', 'build_tracker'] 5 | 6 | 7 | def build_tracker(cfg, use_cuda): 8 | return DeepSort(cfg.DEEPSORT.REID_CKPT, 9 | max_dist=cfg.DEEPSORT.MAX_DIST, min_confidence=cfg.DEEPSORT.MIN_CONFIDENCE, 10 | nms_max_overlap=cfg.DEEPSORT.NMS_MAX_OVERLAP, max_iou_distance=cfg.DEEPSORT.MAX_IOU_DISTANCE, 11 | max_age=cfg.DEEPSORT.MAX_AGE, n_init=cfg.DEEPSORT.N_INIT, nn_budget=cfg.DEEPSORT.NN_BUDGET, use_cuda=use_cuda) 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | -------------------------------------------------------------------------------- /deep_sort/deep_sort/deep/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/deep_sort/deep_sort/deep/__init__.py -------------------------------------------------------------------------------- /deep_sort/deep_sort/deep/checkpoint/.gitkeep: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/deep_sort/deep_sort/deep/checkpoint/.gitkeep -------------------------------------------------------------------------------- /deep_sort/deep_sort/deep/evaluate.py: -------------------------------------------------------------------------------- 1 | import torch 2 | 3 | features = torch.load("features.pth") 4 | qf = features["qf"] 5 | ql = features["ql"] 6 | gf = features["gf"] 7 | gl = features["gl"] 8 | 9 | scores = qf.mm(gf.t()) 10 | res = scores.topk(5, dim=1)[1][:,0] 11 | top1correct = gl[res].eq(ql).sum().item() 12 | 13 | print("Acc top1:{:.3f}".format(top1correct/ql.size(0))) 14 | 15 | 16 | -------------------------------------------------------------------------------- /deep_sort/deep_sort/deep/feature_extractor.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torchvision.transforms as transforms 3 | import numpy as np 4 | import cv2 5 | import logging 6 | 7 | from .model import Net 8 | 9 | class Extractor(object): 10 | def __init__(self, model_path, use_cuda=True): 11 | self.net = Net(reid=True) 12 | self.device = "cuda" if torch.cuda.is_available() and use_cuda else "cpu" 13 | state_dict = torch.load(model_path, map_location=lambda storage, loc: storage)['net_dict'] 14 | self.net.load_state_dict(state_dict) 15 | logger = logging.getLogger("root.tracker") 16 | logger.info("Loading weights from {}... Done!".format(model_path)) 17 | self.net.to(self.device) 18 | self.size = (64, 128) 19 | self.norm = transforms.Compose([ 20 | transforms.ToTensor(), 21 | transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]), 22 | ]) 23 | 24 | 25 | 26 | def _preprocess(self, im_crops): 27 | """ 28 | TODO: 29 | 1. to float with scale from 0 to 1 30 | 2. resize to (64, 128) as Market1501 dataset did 31 | 3. concatenate to a numpy array 32 | 3. to torch Tensor 33 | 4. normalize 34 | """ 35 | def _resize(im, size): 36 | return cv2.resize(im.astype(np.float32)/255., size) 37 | 38 | im_batch = torch.cat([self.norm(_resize(im, self.size)).unsqueeze(0) for im in im_crops], dim=0).float() 39 | return im_batch 40 | 41 | 42 | def __call__(self, im_crops): 43 | im_batch = self._preprocess(im_crops) 44 | with torch.no_grad(): 45 | im_batch = im_batch.to(self.device) 46 | features = self.net(im_batch) 47 | return features.cpu().numpy() 48 | 49 | 50 | if __name__ == '__main__': 51 | img = cv2.imread("demo.jpg")[:,:,(2,1,0)] 52 | extr = Extractor("checkpoint/ckpt.t7") 53 | feature = extr(img) 54 | print(feature.shape) 55 | 56 | -------------------------------------------------------------------------------- /deep_sort/deep_sort/deep/model.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn as nn 3 | import torch.nn.functional as F 4 | 5 | class BasicBlock(nn.Module): 6 | def __init__(self, c_in, c_out,is_downsample=False): 7 | super(BasicBlock,self).__init__() 8 | self.is_downsample = is_downsample 9 | if is_downsample: 10 | self.conv1 = nn.Conv2d(c_in, c_out, 3, stride=2, padding=1, bias=False) 11 | else: 12 | self.conv1 = nn.Conv2d(c_in, c_out, 3, stride=1, padding=1, bias=False) 13 | self.bn1 = nn.BatchNorm2d(c_out) 14 | self.relu = nn.ReLU(True) 15 | self.conv2 = nn.Conv2d(c_out,c_out,3,stride=1,padding=1, bias=False) 16 | self.bn2 = nn.BatchNorm2d(c_out) 17 | if is_downsample: 18 | self.downsample = nn.Sequential( 19 | nn.Conv2d(c_in, c_out, 1, stride=2, bias=False), 20 | nn.BatchNorm2d(c_out) 21 | ) 22 | elif c_in != c_out: 23 | self.downsample = nn.Sequential( 24 | nn.Conv2d(c_in, c_out, 1, stride=1, bias=False), 25 | nn.BatchNorm2d(c_out) 26 | ) 27 | self.is_downsample = True 28 | 29 | def forward(self,x): 30 | y = self.conv1(x) 31 | y = self.bn1(y) 32 | y = self.relu(y) 33 | y = self.conv2(y) 34 | y = self.bn2(y) 35 | if self.is_downsample: 36 | x = self.downsample(x) 37 | return F.relu(x.add(y),True) 38 | 39 | def make_layers(c_in,c_out,repeat_times, is_downsample=False): 40 | blocks = [] 41 | for i in range(repeat_times): 42 | if i ==0: 43 | blocks += [BasicBlock(c_in,c_out, is_downsample=is_downsample),] 44 | else: 45 | blocks += [BasicBlock(c_out,c_out),] 46 | return nn.Sequential(*blocks) 47 | 48 | class Net(nn.Module): 49 | def __init__(self, num_classes=751 ,reid=False): 50 | super(Net,self).__init__() 51 | # 3 128 64 52 | self.conv = nn.Sequential( 53 | nn.Conv2d(3,64,3,stride=1,padding=1), 54 | nn.BatchNorm2d(64), 55 | nn.ReLU(inplace=True), 56 | # nn.Conv2d(32,32,3,stride=1,padding=1), 57 | # nn.BatchNorm2d(32), 58 | # nn.ReLU(inplace=True), 59 | nn.MaxPool2d(3,2,padding=1), 60 | ) 61 | # 32 64 32 62 | self.layer1 = make_layers(64,64,2,False) 63 | # 32 64 32 64 | self.layer2 = make_layers(64,128,2,True) 65 | # 64 32 16 66 | self.layer3 = make_layers(128,256,2,True) 67 | # 128 16 8 68 | self.layer4 = make_layers(256,512,2,True) 69 | # 256 8 4 70 | self.avgpool = nn.AvgPool2d((8,4),1) 71 | # 256 1 1 72 | self.reid = reid 73 | self.classifier = nn.Sequential( 74 | nn.Linear(512, 256), 75 | nn.BatchNorm1d(256), 76 | nn.ReLU(inplace=True), 77 | nn.Dropout(), 78 | nn.Linear(256, num_classes), 79 | ) 80 | 81 | def forward(self, x): 82 | x = self.conv(x) 83 | x = self.layer1(x) 84 | x = self.layer2(x) 85 | x = self.layer3(x) 86 | x = self.layer4(x) 87 | x = self.avgpool(x) 88 | x = x.view(x.size(0),-1) 89 | # B x 128 90 | if self.reid: 91 | x = x.div(x.norm(p=2,dim=1,keepdim=True)) 92 | return x 93 | # classifier 94 | x = self.classifier(x) 95 | return x 96 | 97 | 98 | if __name__ == '__main__': 99 | net = Net() 100 | x = torch.randn(4,3,128,64) 101 | y = net(x) 102 | import ipdb; ipdb.set_trace() 103 | 104 | 105 | -------------------------------------------------------------------------------- /deep_sort/deep_sort/deep/original_model.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn as nn 3 | import torch.nn.functional as F 4 | 5 | class BasicBlock(nn.Module): 6 | def __init__(self, c_in, c_out,is_downsample=False): 7 | super(BasicBlock,self).__init__() 8 | self.is_downsample = is_downsample 9 | if is_downsample: 10 | self.conv1 = nn.Conv2d(c_in, c_out, 3, stride=2, padding=1, bias=False) 11 | else: 12 | self.conv1 = nn.Conv2d(c_in, c_out, 3, stride=1, padding=1, bias=False) 13 | self.bn1 = nn.BatchNorm2d(c_out) 14 | self.relu = nn.ReLU(True) 15 | self.conv2 = nn.Conv2d(c_out,c_out,3,stride=1,padding=1, bias=False) 16 | self.bn2 = nn.BatchNorm2d(c_out) 17 | if is_downsample: 18 | self.downsample = nn.Sequential( 19 | nn.Conv2d(c_in, c_out, 1, stride=2, bias=False), 20 | nn.BatchNorm2d(c_out) 21 | ) 22 | elif c_in != c_out: 23 | self.downsample = nn.Sequential( 24 | nn.Conv2d(c_in, c_out, 1, stride=1, bias=False), 25 | nn.BatchNorm2d(c_out) 26 | ) 27 | self.is_downsample = True 28 | 29 | def forward(self,x): 30 | y = self.conv1(x) 31 | y = self.bn1(y) 32 | y = self.relu(y) 33 | y = self.conv2(y) 34 | y = self.bn2(y) 35 | if self.is_downsample: 36 | x = self.downsample(x) 37 | return F.relu(x.add(y),True) 38 | 39 | def make_layers(c_in,c_out,repeat_times, is_downsample=False): 40 | blocks = [] 41 | for i in range(repeat_times): 42 | if i ==0: 43 | blocks += [BasicBlock(c_in,c_out, is_downsample=is_downsample),] 44 | else: 45 | blocks += [BasicBlock(c_out,c_out),] 46 | return nn.Sequential(*blocks) 47 | 48 | class Net(nn.Module): 49 | def __init__(self, num_classes=625 ,reid=False): 50 | super(Net,self).__init__() 51 | # 3 128 64 52 | self.conv = nn.Sequential( 53 | nn.Conv2d(3,32,3,stride=1,padding=1), 54 | nn.BatchNorm2d(32), 55 | nn.ELU(inplace=True), 56 | nn.Conv2d(32,32,3,stride=1,padding=1), 57 | nn.BatchNorm2d(32), 58 | nn.ELU(inplace=True), 59 | nn.MaxPool2d(3,2,padding=1), 60 | ) 61 | # 32 64 32 62 | self.layer1 = make_layers(32,32,2,False) 63 | # 32 64 32 64 | self.layer2 = make_layers(32,64,2,True) 65 | # 64 32 16 66 | self.layer3 = make_layers(64,128,2,True) 67 | # 128 16 8 68 | self.dense = nn.Sequential( 69 | nn.Dropout(p=0.6), 70 | nn.Linear(128*16*8, 128), 71 | nn.BatchNorm1d(128), 72 | nn.ELU(inplace=True) 73 | ) 74 | # 256 1 1 75 | self.reid = reid 76 | self.batch_norm = nn.BatchNorm1d(128) 77 | self.classifier = nn.Sequential( 78 | nn.Linear(128, num_classes), 79 | ) 80 | 81 | def forward(self, x): 82 | x = self.conv(x) 83 | x = self.layer1(x) 84 | x = self.layer2(x) 85 | x = self.layer3(x) 86 | 87 | x = x.view(x.size(0),-1) 88 | if self.reid: 89 | x = self.dense[0](x) 90 | x = self.dense[1](x) 91 | x = x.div(x.norm(p=2,dim=1,keepdim=True)) 92 | return x 93 | x = self.dense(x) 94 | # B x 128 95 | # classifier 96 | x = self.classifier(x) 97 | return x 98 | 99 | 100 | if __name__ == '__main__': 101 | net = Net(reid=True) 102 | x = torch.randn(4,3,128,64) 103 | y = net(x) 104 | import ipdb; ipdb.set_trace() 105 | 106 | 107 | -------------------------------------------------------------------------------- /deep_sort/deep_sort/deep/test.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.backends.cudnn as cudnn 3 | import torchvision 4 | 5 | import argparse 6 | import os 7 | 8 | from model import Net 9 | 10 | parser = argparse.ArgumentParser(description="Train on market1501") 11 | parser.add_argument("--data-dir",default='data',type=str) 12 | parser.add_argument("--no-cuda",action="store_true") 13 | parser.add_argument("--gpu-id",default=0,type=int) 14 | args = parser.parse_args() 15 | 16 | # device 17 | device = "cuda:{}".format(args.gpu_id) if torch.cuda.is_available() and not args.no_cuda else "cpu" 18 | if torch.cuda.is_available() and not args.no_cuda: 19 | cudnn.benchmark = True 20 | 21 | # data loader 22 | root = args.data_dir 23 | query_dir = os.path.join(root,"query") 24 | gallery_dir = os.path.join(root,"gallery") 25 | transform = torchvision.transforms.Compose([ 26 | torchvision.transforms.Resize((128,64)), 27 | torchvision.transforms.ToTensor(), 28 | torchvision.transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) 29 | ]) 30 | queryloader = torch.utils.data.DataLoader( 31 | torchvision.datasets.ImageFolder(query_dir, transform=transform), 32 | batch_size=64, shuffle=False 33 | ) 34 | galleryloader = torch.utils.data.DataLoader( 35 | torchvision.datasets.ImageFolder(gallery_dir, transform=transform), 36 | batch_size=64, shuffle=False 37 | ) 38 | 39 | # net definition 40 | net = Net(reid=True) 41 | assert os.path.isfile("./checkpoint/ckpt.t7"), "Error: no checkpoint file found!" 42 | print('Loading from checkpoint/ckpt.t7') 43 | checkpoint = torch.load("./checkpoint/ckpt.t7") 44 | net_dict = checkpoint['net_dict'] 45 | net.load_state_dict(net_dict, strict=False) 46 | net.eval() 47 | net.to(device) 48 | 49 | # compute features 50 | query_features = torch.tensor([]).float() 51 | query_labels = torch.tensor([]).long() 52 | gallery_features = torch.tensor([]).float() 53 | gallery_labels = torch.tensor([]).long() 54 | 55 | with torch.no_grad(): 56 | for idx,(inputs,labels) in enumerate(queryloader): 57 | inputs = inputs.to(device) 58 | features = net(inputs).cpu() 59 | query_features = torch.cat((query_features, features), dim=0) 60 | query_labels = torch.cat((query_labels, labels)) 61 | 62 | for idx,(inputs,labels) in enumerate(galleryloader): 63 | inputs = inputs.to(device) 64 | features = net(inputs).cpu() 65 | gallery_features = torch.cat((gallery_features, features), dim=0) 66 | gallery_labels = torch.cat((gallery_labels, labels)) 67 | 68 | gallery_labels -= 2 69 | 70 | # save features 71 | features = { 72 | "qf": query_features, 73 | "ql": query_labels, 74 | "gf": gallery_features, 75 | "gl": gallery_labels 76 | } 77 | torch.save(features,"features.pth") -------------------------------------------------------------------------------- /deep_sort/deep_sort/deep/train.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/deep_sort/deep_sort/deep/train.jpg -------------------------------------------------------------------------------- /deep_sort/deep_sort/deep_sort.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import torch 3 | 4 | from .deep.feature_extractor import Extractor 5 | from .sort.nn_matching import NearestNeighborDistanceMetric 6 | from .sort.preprocessing import non_max_suppression 7 | from .sort.detection import Detection 8 | from .sort.tracker import Tracker 9 | 10 | 11 | __all__ = ['DeepSort'] 12 | 13 | 14 | class DeepSort(object): 15 | def __init__(self, model_path, max_dist=0.2, min_confidence=0.3, nms_max_overlap=1.0, max_iou_distance=0.7, max_age=70, n_init=3, nn_budget=100, use_cuda=True): 16 | self.min_confidence = min_confidence 17 | self.nms_max_overlap = nms_max_overlap 18 | 19 | self.extractor = Extractor(model_path, use_cuda=use_cuda) 20 | 21 | max_cosine_distance = max_dist 22 | nn_budget = 100 23 | metric = NearestNeighborDistanceMetric("cosine", max_cosine_distance, nn_budget) 24 | self.tracker = Tracker(metric, max_iou_distance=max_iou_distance, max_age=max_age, n_init=n_init) 25 | 26 | def update(self, bbox_xywh, confidences, ori_img): 27 | self.height, self.width = ori_img.shape[:2] 28 | # generate detections 29 | features = self._get_features(bbox_xywh, ori_img) 30 | bbox_tlwh = self._xywh_to_tlwh(bbox_xywh) 31 | detections = [Detection(bbox_tlwh[i], conf, features[i]) for i,conf in enumerate(confidences) if conf>self.min_confidence] 32 | 33 | # run on non-maximum supression 34 | boxes = np.array([d.tlwh for d in detections]) 35 | scores = np.array([d.confidence for d in detections]) 36 | indices = non_max_suppression(boxes, self.nms_max_overlap, scores) 37 | detections = [detections[i] for i in indices] 38 | 39 | # update tracker 40 | self.tracker.predict() 41 | self.tracker.update(detections) 42 | 43 | # output bbox identities 44 | outputs = [] 45 | for track in self.tracker.tracks: 46 | if not track.is_confirmed() or track.time_since_update > 1: 47 | continue 48 | box = track.to_tlwh() 49 | x1,y1,x2,y2 = self._tlwh_to_xyxy(box) 50 | track_id = track.track_id 51 | outputs.append(np.array([x1,y1,x2,y2,track_id], dtype=np.int)) 52 | if len(outputs) > 0: 53 | outputs = np.stack(outputs,axis=0) 54 | return outputs 55 | 56 | 57 | """ 58 | TODO: 59 | Convert bbox from xc_yc_w_h to xtl_ytl_w_h 60 | Thanks JieChen91@github.com for reporting this bug! 61 | """ 62 | @staticmethod 63 | def _xywh_to_tlwh(bbox_xywh): 64 | if isinstance(bbox_xywh, np.ndarray): 65 | bbox_tlwh = bbox_xywh.copy() 66 | elif isinstance(bbox_xywh, torch.Tensor): 67 | bbox_tlwh = bbox_xywh.clone() 68 | bbox_tlwh[:,0] = bbox_xywh[:,0] - bbox_xywh[:,2]/2. 69 | bbox_tlwh[:,1] = bbox_xywh[:,1] - bbox_xywh[:,3]/2. 70 | return bbox_tlwh 71 | 72 | 73 | def _xywh_to_xyxy(self, bbox_xywh): 74 | x,y,w,h = bbox_xywh 75 | x1 = max(int(x-w/2),0) 76 | x2 = min(int(x+w/2),self.width-1) 77 | y1 = max(int(y-h/2),0) 78 | y2 = min(int(y+h/2),self.height-1) 79 | return x1,y1,x2,y2 80 | 81 | def _tlwh_to_xyxy(self, bbox_tlwh): 82 | """ 83 | TODO: 84 | Convert bbox from xtl_ytl_w_h to xc_yc_w_h 85 | Thanks JieChen91@github.com for reporting this bug! 86 | """ 87 | x,y,w,h = bbox_tlwh 88 | x1 = max(int(x),0) 89 | x2 = min(int(x+w),self.width-1) 90 | y1 = max(int(y),0) 91 | y2 = min(int(y+h),self.height-1) 92 | return x1,y1,x2,y2 93 | 94 | def _xyxy_to_tlwh(self, bbox_xyxy): 95 | x1,y1,x2,y2 = bbox_xyxy 96 | 97 | t = x1 98 | l = y1 99 | w = int(x2-x1) 100 | h = int(y2-y1) 101 | return t,l,w,h 102 | 103 | def _get_features(self, bbox_xywh, ori_img): 104 | im_crops = [] 105 | for box in bbox_xywh: 106 | x1,y1,x2,y2 = self._xywh_to_xyxy(box) 107 | im = ori_img[y1:y2,x1:x2] 108 | im_crops.append(im) 109 | if im_crops: 110 | features = self.extractor(im_crops) 111 | else: 112 | features = np.array([]) 113 | return features 114 | 115 | 116 | -------------------------------------------------------------------------------- /deep_sort/deep_sort/sort/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/deep_sort/deep_sort/sort/__init__.py -------------------------------------------------------------------------------- /deep_sort/deep_sort/sort/detection.py: -------------------------------------------------------------------------------- 1 | # vim: expandtab:ts=4:sw=4 2 | import numpy as np 3 | 4 | 5 | class Detection(object): 6 | """ 7 | This class represents a bounding box detection in a single image. 8 | 9 | Parameters 10 | ---------- 11 | tlwh : array_like 12 | Bounding box in format `(x, y, w, h)`. 13 | confidence : float 14 | Detector confidence score. 15 | feature : array_like 16 | A feature vector that describes the object contained in this image. 17 | 18 | Attributes 19 | ---------- 20 | tlwh : ndarray 21 | Bounding box in format `(top left x, top left y, width, height)`. 22 | confidence : ndarray 23 | Detector confidence score. 24 | feature : ndarray | NoneType 25 | A feature vector that describes the object contained in this image. 26 | 27 | """ 28 | 29 | def __init__(self, tlwh, confidence, feature): 30 | self.tlwh = np.asarray(tlwh, dtype=np.float) 31 | self.confidence = float(confidence) 32 | self.feature = np.asarray(feature, dtype=np.float32) 33 | 34 | def to_tlbr(self): 35 | """Convert bounding box to format `(min x, min y, max x, max y)`, i.e., 36 | `(top left, bottom right)`. 37 | """ 38 | ret = self.tlwh.copy() 39 | ret[2:] += ret[:2] 40 | return ret 41 | 42 | def to_xyah(self): 43 | """Convert bounding box to format `(center x, center y, aspect ratio, 44 | height)`, where the aspect ratio is `width / height`. 45 | """ 46 | ret = self.tlwh.copy() 47 | ret[:2] += ret[2:] / 2 48 | ret[2] /= ret[3] 49 | return ret 50 | -------------------------------------------------------------------------------- /deep_sort/deep_sort/sort/iou_matching.py: -------------------------------------------------------------------------------- 1 | # vim: expandtab:ts=4:sw=4 2 | from __future__ import absolute_import 3 | import numpy as np 4 | from . import linear_assignment 5 | 6 | 7 | def iou(bbox, candidates): 8 | """Computer intersection over union. 9 | 10 | Parameters 11 | ---------- 12 | bbox : ndarray 13 | A bounding box in format `(top left x, top left y, width, height)`. 14 | candidates : ndarray 15 | A matrix of candidate bounding boxes (one per row) in the same format 16 | as `bbox`. 17 | 18 | Returns 19 | ------- 20 | ndarray 21 | The intersection over union in [0, 1] between the `bbox` and each 22 | candidate. A higher score means a larger fraction of the `bbox` is 23 | occluded by the candidate. 24 | 25 | """ 26 | bbox_tl, bbox_br = bbox[:2], bbox[:2] + bbox[2:] 27 | candidates_tl = candidates[:, :2] 28 | candidates_br = candidates[:, :2] + candidates[:, 2:] 29 | 30 | tl = np.c_[np.maximum(bbox_tl[0], candidates_tl[:, 0])[:, np.newaxis], 31 | np.maximum(bbox_tl[1], candidates_tl[:, 1])[:, np.newaxis]] 32 | br = np.c_[np.minimum(bbox_br[0], candidates_br[:, 0])[:, np.newaxis], 33 | np.minimum(bbox_br[1], candidates_br[:, 1])[:, np.newaxis]] 34 | wh = np.maximum(0., br - tl) 35 | 36 | area_intersection = wh.prod(axis=1) 37 | area_bbox = bbox[2:].prod() 38 | area_candidates = candidates[:, 2:].prod(axis=1) 39 | return area_intersection / (area_bbox + area_candidates - area_intersection) 40 | 41 | 42 | def iou_cost(tracks, detections, track_indices=None, 43 | detection_indices=None): 44 | """An intersection over union distance metric. 45 | 46 | Parameters 47 | ---------- 48 | tracks : List[deep_sort.track.Track] 49 | A list of tracks. 50 | detections : List[deep_sort.detection.Detection] 51 | A list of detections. 52 | track_indices : Optional[List[int]] 53 | A list of indices to tracks that should be matched. Defaults to 54 | all `tracks`. 55 | detection_indices : Optional[List[int]] 56 | A list of indices to detections that should be matched. Defaults 57 | to all `detections`. 58 | 59 | Returns 60 | ------- 61 | ndarray 62 | Returns a cost matrix of shape 63 | len(track_indices), len(detection_indices) where entry (i, j) is 64 | `1 - iou(tracks[track_indices[i]], detections[detection_indices[j]])`. 65 | 66 | """ 67 | if track_indices is None: 68 | track_indices = np.arange(len(tracks)) 69 | if detection_indices is None: 70 | detection_indices = np.arange(len(detections)) 71 | 72 | cost_matrix = np.zeros((len(track_indices), len(detection_indices))) 73 | for row, track_idx in enumerate(track_indices): 74 | if tracks[track_idx].time_since_update > 1: 75 | cost_matrix[row, :] = linear_assignment.INFTY_COST 76 | continue 77 | 78 | bbox = tracks[track_idx].to_tlwh() 79 | candidates = np.asarray([detections[i].tlwh for i in detection_indices]) 80 | cost_matrix[row, :] = 1. - iou(bbox, candidates) 81 | return cost_matrix 82 | -------------------------------------------------------------------------------- /deep_sort/deep_sort/sort/nn_matching.py: -------------------------------------------------------------------------------- 1 | # vim: expandtab:ts=4:sw=4 2 | import numpy as np 3 | 4 | 5 | def _pdist(a, b): 6 | """Compute pair-wise squared distance between points in `a` and `b`. 7 | 8 | Parameters 9 | ---------- 10 | a : array_like 11 | An NxM matrix of N samples of dimensionality M. 12 | b : array_like 13 | An LxM matrix of L samples of dimensionality M. 14 | 15 | Returns 16 | ------- 17 | ndarray 18 | Returns a matrix of size len(a), len(b) such that eleement (i, j) 19 | contains the squared distance between `a[i]` and `b[j]`. 20 | 21 | """ 22 | a, b = np.asarray(a), np.asarray(b) 23 | if len(a) == 0 or len(b) == 0: 24 | return np.zeros((len(a), len(b))) 25 | a2, b2 = np.square(a).sum(axis=1), np.square(b).sum(axis=1) 26 | r2 = -2. * np.dot(a, b.T) + a2[:, None] + b2[None, :] 27 | r2 = np.clip(r2, 0., float(np.inf)) 28 | return r2 29 | 30 | 31 | def _cosine_distance(a, b, data_is_normalized=False): 32 | """Compute pair-wise cosine distance between points in `a` and `b`. 33 | 34 | Parameters 35 | ---------- 36 | a : array_like 37 | An NxM matrix of N samples of dimensionality M. 38 | b : array_like 39 | An LxM matrix of L samples of dimensionality M. 40 | data_is_normalized : Optional[bool] 41 | If True, assumes rows in a and b are unit length vectors. 42 | Otherwise, a and b are explicitly normalized to lenght 1. 43 | 44 | Returns 45 | ------- 46 | ndarray 47 | Returns a matrix of size len(a), len(b) such that eleement (i, j) 48 | contains the squared distance between `a[i]` and `b[j]`. 49 | 50 | """ 51 | if not data_is_normalized: 52 | a = np.asarray(a) / np.linalg.norm(a, axis=1, keepdims=True) 53 | b = np.asarray(b) / np.linalg.norm(b, axis=1, keepdims=True) 54 | return 1. - np.dot(a, b.T) 55 | 56 | 57 | def _nn_euclidean_distance(x, y): 58 | """ Helper function for nearest neighbor distance metric (Euclidean). 59 | 60 | Parameters 61 | ---------- 62 | x : ndarray 63 | A matrix of N row-vectors (sample points). 64 | y : ndarray 65 | A matrix of M row-vectors (query points). 66 | 67 | Returns 68 | ------- 69 | ndarray 70 | A vector of length M that contains for each entry in `y` the 71 | smallest Euclidean distance to a sample in `x`. 72 | 73 | """ 74 | distances = _pdist(x, y) 75 | return np.maximum(0.0, distances.min(axis=0)) 76 | 77 | 78 | def _nn_cosine_distance(x, y): 79 | """ Helper function for nearest neighbor distance metric (cosine). 80 | 81 | Parameters 82 | ---------- 83 | x : ndarray 84 | A matrix of N row-vectors (sample points). 85 | y : ndarray 86 | A matrix of M row-vectors (query points). 87 | 88 | Returns 89 | ------- 90 | ndarray 91 | A vector of length M that contains for each entry in `y` the 92 | smallest cosine distance to a sample in `x`. 93 | 94 | """ 95 | distances = _cosine_distance(x, y) 96 | return distances.min(axis=0) 97 | 98 | 99 | class NearestNeighborDistanceMetric(object): 100 | """ 101 | A nearest neighbor distance metric that, for each target, returns 102 | the closest distance to any sample that has been observed so far. 103 | 104 | Parameters 105 | ---------- 106 | metric : str 107 | Either "euclidean" or "cosine". 108 | matching_threshold: float 109 | The matching threshold. Samples with larger distance are considered an 110 | invalid match. 111 | budget : Optional[int] 112 | If not None, fix samples per class to at most this number. Removes 113 | the oldest samples when the budget is reached. 114 | 115 | Attributes 116 | ---------- 117 | samples : Dict[int -> List[ndarray]] 118 | A dictionary that maps from target identities to the list of samples 119 | that have been observed so far. 120 | 121 | """ 122 | 123 | def __init__(self, metric, matching_threshold, budget=None): 124 | 125 | 126 | if metric == "euclidean": 127 | self._metric = _nn_euclidean_distance 128 | elif metric == "cosine": 129 | self._metric = _nn_cosine_distance 130 | else: 131 | raise ValueError( 132 | "Invalid metric; must be either 'euclidean' or 'cosine'") 133 | self.matching_threshold = matching_threshold 134 | self.budget = budget 135 | self.samples = {} 136 | 137 | def partial_fit(self, features, targets, active_targets): 138 | """Update the distance metric with new data. 139 | 140 | Parameters 141 | ---------- 142 | features : ndarray 143 | An NxM matrix of N features of dimensionality M. 144 | targets : ndarray 145 | An integer array of associated target identities. 146 | active_targets : List[int] 147 | A list of targets that are currently present in the scene. 148 | 149 | """ 150 | for feature, target in zip(features, targets): 151 | self.samples.setdefault(target, []).append(feature) 152 | if self.budget is not None: 153 | self.samples[target] = self.samples[target][-self.budget:] 154 | self.samples = {k: self.samples[k] for k in active_targets} 155 | 156 | def distance(self, features, targets): 157 | """Compute distance between features and targets. 158 | 159 | Parameters 160 | ---------- 161 | features : ndarray 162 | An NxM matrix of N features of dimensionality M. 163 | targets : List[int] 164 | A list of targets to match the given `features` against. 165 | 166 | Returns 167 | ------- 168 | ndarray 169 | Returns a cost matrix of shape len(targets), len(features), where 170 | element (i, j) contains the closest squared distance between 171 | `targets[i]` and `features[j]`. 172 | 173 | """ 174 | cost_matrix = np.zeros((len(targets), len(features))) 175 | for i, target in enumerate(targets): 176 | cost_matrix[i, :] = self._metric(self.samples[target], features) 177 | return cost_matrix 178 | -------------------------------------------------------------------------------- /deep_sort/deep_sort/sort/preprocessing.py: -------------------------------------------------------------------------------- 1 | # vim: expandtab:ts=4:sw=4 2 | import numpy as np 3 | import cv2 4 | 5 | 6 | def non_max_suppression(boxes, max_bbox_overlap, scores=None): 7 | """Suppress overlapping detections. 8 | 9 | Original code from [1]_ has been adapted to include confidence score. 10 | 11 | .. [1] http://www.pyimagesearch.com/2015/02/16/ 12 | faster-non-maximum-suppression-python/ 13 | 14 | Examples 15 | -------- 16 | 17 | >>> boxes = [d.roi for d in detections] 18 | >>> scores = [d.confidence for d in detections] 19 | >>> indices = non_max_suppression(boxes, max_bbox_overlap, scores) 20 | >>> detections = [detections[i] for i in indices] 21 | 22 | Parameters 23 | ---------- 24 | boxes : ndarray 25 | Array of ROIs (x, y, width, height). 26 | max_bbox_overlap : float 27 | ROIs that overlap more than this values are suppressed. 28 | scores : Optional[array_like] 29 | Detector confidence score. 30 | 31 | Returns 32 | ------- 33 | List[int] 34 | Returns indices of detections that have survived non-maxima suppression. 35 | 36 | """ 37 | if len(boxes) == 0: 38 | return [] 39 | 40 | boxes = boxes.astype(np.float) 41 | pick = [] 42 | 43 | x1 = boxes[:, 0] 44 | y1 = boxes[:, 1] 45 | x2 = boxes[:, 2] + boxes[:, 0] 46 | y2 = boxes[:, 3] + boxes[:, 1] 47 | 48 | area = (x2 - x1 + 1) * (y2 - y1 + 1) 49 | if scores is not None: 50 | idxs = np.argsort(scores) 51 | else: 52 | idxs = np.argsort(y2) 53 | 54 | while len(idxs) > 0: 55 | last = len(idxs) - 1 56 | i = idxs[last] 57 | pick.append(i) 58 | 59 | xx1 = np.maximum(x1[i], x1[idxs[:last]]) 60 | yy1 = np.maximum(y1[i], y1[idxs[:last]]) 61 | xx2 = np.minimum(x2[i], x2[idxs[:last]]) 62 | yy2 = np.minimum(y2[i], y2[idxs[:last]]) 63 | 64 | w = np.maximum(0, xx2 - xx1 + 1) 65 | h = np.maximum(0, yy2 - yy1 + 1) 66 | 67 | overlap = (w * h) / area[idxs[:last]] 68 | 69 | idxs = np.delete( 70 | idxs, np.concatenate( 71 | ([last], np.where(overlap > max_bbox_overlap)[0]))) 72 | 73 | return pick 74 | -------------------------------------------------------------------------------- /deep_sort/deep_sort/sort/track.py: -------------------------------------------------------------------------------- 1 | # vim: expandtab:ts=4:sw=4 2 | 3 | 4 | class TrackState: 5 | """ 6 | Enumeration type for the single target track state. Newly created tracks are 7 | classified as `tentative` until enough evidence has been collected. Then, 8 | the track state is changed to `confirmed`. Tracks that are no longer alive 9 | are classified as `deleted` to mark them for removal from the set of active 10 | tracks. 11 | 12 | """ 13 | 14 | Tentative = 1 15 | Confirmed = 2 16 | Deleted = 3 17 | 18 | 19 | class Track: 20 | """ 21 | A single target track with state space `(x, y, a, h)` and associated 22 | velocities, where `(x, y)` is the center of the bounding box, `a` is the 23 | aspect ratio and `h` is the height. 24 | 25 | Parameters 26 | ---------- 27 | mean : ndarray 28 | Mean vector of the initial state distribution. 29 | covariance : ndarray 30 | Covariance matrix of the initial state distribution. 31 | track_id : int 32 | A unique track identifier. 33 | n_init : int 34 | Number of consecutive detections before the track is confirmed. The 35 | track state is set to `Deleted` if a miss occurs within the first 36 | `n_init` frames. 37 | max_age : int 38 | The maximum number of consecutive misses before the track state is 39 | set to `Deleted`. 40 | feature : Optional[ndarray] 41 | Feature vector of the detection this track originates from. If not None, 42 | this feature is added to the `features` cache. 43 | 44 | Attributes 45 | ---------- 46 | mean : ndarray 47 | Mean vector of the initial state distribution. 48 | covariance : ndarray 49 | Covariance matrix of the initial state distribution. 50 | track_id : int 51 | A unique track identifier. 52 | hits : int 53 | Total number of measurement updates. 54 | age : int 55 | Total number of frames since first occurance. 56 | time_since_update : int 57 | Total number of frames since last measurement update. 58 | state : TrackState 59 | The current track state. 60 | features : List[ndarray] 61 | A cache of features. On each measurement update, the associated feature 62 | vector is added to this list. 63 | 64 | """ 65 | 66 | def __init__(self, mean, covariance, track_id, n_init, max_age, 67 | feature=None): 68 | self.mean = mean 69 | self.covariance = covariance 70 | self.track_id = track_id 71 | self.hits = 1 72 | self.age = 1 73 | self.time_since_update = 0 74 | 75 | self.state = TrackState.Tentative 76 | self.features = [] 77 | if feature is not None: 78 | self.features.append(feature) 79 | 80 | self._n_init = n_init 81 | self._max_age = max_age 82 | 83 | def to_tlwh(self): 84 | """Get current position in bounding box format `(top left x, top left y, 85 | width, height)`. 86 | 87 | Returns 88 | ------- 89 | ndarray 90 | The bounding box. 91 | 92 | """ 93 | ret = self.mean[:4].copy() 94 | ret[2] *= ret[3] 95 | ret[:2] -= ret[2:] / 2 96 | return ret 97 | 98 | def to_tlbr(self): 99 | """Get current position in bounding box format `(min x, miny, max x, 100 | max y)`. 101 | 102 | Returns 103 | ------- 104 | ndarray 105 | The bounding box. 106 | 107 | """ 108 | ret = self.to_tlwh() 109 | ret[2:] = ret[:2] + ret[2:] 110 | return ret 111 | 112 | def predict(self, kf): 113 | """Propagate the state distribution to the current time step using a 114 | Kalman filter prediction step. 115 | 116 | Parameters 117 | ---------- 118 | kf : kalman_filter.KalmanFilter 119 | The Kalman filter. 120 | 121 | """ 122 | self.mean, self.covariance = kf.predict(self.mean, self.covariance) 123 | self.age += 1 124 | self.time_since_update += 1 125 | 126 | def update(self, kf, detection): 127 | """Perform Kalman filter measurement update step and update the feature 128 | cache. 129 | 130 | Parameters 131 | ---------- 132 | kf : kalman_filter.KalmanFilter 133 | The Kalman filter. 134 | detection : Detection 135 | The associated detection. 136 | 137 | """ 138 | self.mean, self.covariance = kf.update( 139 | self.mean, self.covariance, detection.to_xyah()) 140 | self.features.append(detection.feature) 141 | 142 | self.hits += 1 143 | self.time_since_update = 0 144 | if self.state == TrackState.Tentative and self.hits >= self._n_init: 145 | self.state = TrackState.Confirmed 146 | 147 | def mark_missed(self): 148 | """Mark this track as missed (no association at the current time step). 149 | """ 150 | if self.state == TrackState.Tentative: 151 | self.state = TrackState.Deleted 152 | elif self.time_since_update > self._max_age: 153 | self.state = TrackState.Deleted 154 | 155 | def is_tentative(self): 156 | """Returns True if this track is tentative (unconfirmed). 157 | """ 158 | return self.state == TrackState.Tentative 159 | 160 | def is_confirmed(self): 161 | """Returns True if this track is confirmed.""" 162 | return self.state == TrackState.Confirmed 163 | 164 | def is_deleted(self): 165 | """Returns True if this track is dead and should be deleted.""" 166 | return self.state == TrackState.Deleted 167 | -------------------------------------------------------------------------------- /deep_sort/deep_sort/sort/tracker.py: -------------------------------------------------------------------------------- 1 | # vim: expandtab:ts=4:sw=4 2 | from __future__ import absolute_import 3 | import numpy as np 4 | from . import kalman_filter 5 | from . import linear_assignment 6 | from . import iou_matching 7 | from .track import Track 8 | 9 | 10 | class Tracker: 11 | """ 12 | This is the multi-target tracker. 13 | 14 | Parameters 15 | ---------- 16 | metric : nn_matching.NearestNeighborDistanceMetric 17 | A distance metric for measurement-to-track association. 18 | max_age : int 19 | Maximum number of missed misses before a track is deleted. 20 | n_init : int 21 | Number of consecutive detections before the track is confirmed. The 22 | track state is set to `Deleted` if a miss occurs within the first 23 | `n_init` frames. 24 | 25 | Attributes 26 | ---------- 27 | metric : nn_matching.NearestNeighborDistanceMetric 28 | The distance metric used for measurement to track association. 29 | max_age : int 30 | Maximum number of missed misses before a track is deleted. 31 | n_init : int 32 | Number of frames that a track remains in initialization phase. 33 | kf : kalman_filter.KalmanFilter 34 | A Kalman filter to filter target trajectories in image space. 35 | tracks : List[Track] 36 | The list of active tracks at the current time step. 37 | 38 | """ 39 | 40 | def __init__(self, metric, max_iou_distance=0.7, max_age=70, n_init=3): 41 | self.metric = metric 42 | self.max_iou_distance = max_iou_distance 43 | self.max_age = max_age 44 | self.n_init = n_init 45 | 46 | self.kf = kalman_filter.KalmanFilter() 47 | self.tracks = [] 48 | self._next_id = 1 49 | 50 | def predict(self): 51 | """Propagate track state distributions one time step forward. 52 | 53 | This function should be called once every time step, before `update`. 54 | """ 55 | for track in self.tracks: 56 | track.predict(self.kf) 57 | 58 | def update(self, detections): 59 | """Perform measurement update and track management. 60 | 61 | Parameters 62 | ---------- 63 | detections : List[deep_sort.detection.Detection] 64 | A list of detections at the current time step. 65 | 66 | """ 67 | # Run matching cascade. 68 | matches, unmatched_tracks, unmatched_detections = \ 69 | self._match(detections) 70 | 71 | # Update track set. 72 | for track_idx, detection_idx in matches: 73 | self.tracks[track_idx].update( 74 | self.kf, detections[detection_idx]) 75 | for track_idx in unmatched_tracks: 76 | self.tracks[track_idx].mark_missed() 77 | for detection_idx in unmatched_detections: 78 | self._initiate_track(detections[detection_idx]) 79 | self.tracks = [t for t in self.tracks if not t.is_deleted()] 80 | 81 | # Update distance metric. 82 | active_targets = [t.track_id for t in self.tracks if t.is_confirmed()] 83 | features, targets = [], [] 84 | for track in self.tracks: 85 | if not track.is_confirmed(): 86 | continue 87 | features += track.features 88 | targets += [track.track_id for _ in track.features] 89 | track.features = [] 90 | self.metric.partial_fit( 91 | np.asarray(features), np.asarray(targets), active_targets) 92 | 93 | def _match(self, detections): 94 | 95 | def gated_metric(tracks, dets, track_indices, detection_indices): 96 | features = np.array([dets[i].feature for i in detection_indices]) 97 | targets = np.array([tracks[i].track_id for i in track_indices]) 98 | cost_matrix = self.metric.distance(features, targets) 99 | cost_matrix = linear_assignment.gate_cost_matrix( 100 | self.kf, cost_matrix, tracks, dets, track_indices, 101 | detection_indices) 102 | 103 | return cost_matrix 104 | 105 | # Split track set into confirmed and unconfirmed tracks. 106 | confirmed_tracks = [ 107 | i for i, t in enumerate(self.tracks) if t.is_confirmed()] 108 | unconfirmed_tracks = [ 109 | i for i, t in enumerate(self.tracks) if not t.is_confirmed()] 110 | 111 | # Associate confirmed tracks using appearance features. 112 | matches_a, unmatched_tracks_a, unmatched_detections = \ 113 | linear_assignment.matching_cascade( 114 | gated_metric, self.metric.matching_threshold, self.max_age, 115 | self.tracks, detections, confirmed_tracks) 116 | 117 | # Associate remaining tracks together with unconfirmed tracks using IOU. 118 | iou_track_candidates = unconfirmed_tracks + [ 119 | k for k in unmatched_tracks_a if 120 | self.tracks[k].time_since_update == 1] 121 | unmatched_tracks_a = [ 122 | k for k in unmatched_tracks_a if 123 | self.tracks[k].time_since_update != 1] 124 | matches_b, unmatched_tracks_b, unmatched_detections = \ 125 | linear_assignment.min_cost_matching( 126 | iou_matching.iou_cost, self.max_iou_distance, self.tracks, 127 | detections, iou_track_candidates, unmatched_detections) 128 | 129 | matches = matches_a + matches_b 130 | unmatched_tracks = list(set(unmatched_tracks_a + unmatched_tracks_b)) 131 | return matches, unmatched_tracks, unmatched_detections 132 | 133 | def _initiate_track(self, detection): 134 | mean, covariance = self.kf.initiate(detection.to_xyah()) 135 | self.tracks.append(Track( 136 | mean, covariance, self._next_id, self.n_init, self.max_age, 137 | detection.feature)) 138 | self._next_id += 1 139 | -------------------------------------------------------------------------------- /deep_sort/demo/1.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/deep_sort/demo/1.jpg -------------------------------------------------------------------------------- /deep_sort/demo/2.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/deep_sort/demo/2.jpg -------------------------------------------------------------------------------- /deep_sort/demo/demo.gif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/deep_sort/demo/demo.gif -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/README.md: -------------------------------------------------------------------------------- 1 | # YOLOv3 for detection 2 | 3 | This is an implemention of YOLOv3 with only the forward part. 4 | 5 | If you want to train YOLOv3 on your custom dataset, please search `YOLOv3` on github. 6 | 7 | ## Quick forward 8 | ```bash 9 | cd YOLOv3 10 | python 11 | ``` -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/__init__.py: -------------------------------------------------------------------------------- 1 | import sys 2 | sys.path.append("detector/YOLOv3") 3 | 4 | 5 | from .detector import YOLOv3 6 | __all__ = ['YOLOv3'] 7 | 8 | 9 | 10 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/cfg/coco.data: -------------------------------------------------------------------------------- 1 | train = coco_train.txt 2 | valid = coco_test.txt 3 | names = data/coco.names 4 | backup = backup 5 | gpus = 0,1,2,3 6 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/cfg/coco.names: -------------------------------------------------------------------------------- 1 | person 2 | bicycle 3 | car 4 | motorbike 5 | aeroplane 6 | bus 7 | train 8 | truck 9 | boat 10 | traffic light 11 | fire hydrant 12 | stop sign 13 | parking meter 14 | bench 15 | bird 16 | cat 17 | dog 18 | horse 19 | sheep 20 | cow 21 | elephant 22 | bear 23 | zebra 24 | giraffe 25 | backpack 26 | umbrella 27 | handbag 28 | tie 29 | suitcase 30 | frisbee 31 | skis 32 | snowboard 33 | sports ball 34 | kite 35 | baseball bat 36 | baseball glove 37 | skateboard 38 | surfboard 39 | tennis racket 40 | bottle 41 | wine glass 42 | cup 43 | fork 44 | knife 45 | spoon 46 | bowl 47 | banana 48 | apple 49 | sandwich 50 | orange 51 | broccoli 52 | carrot 53 | hot dog 54 | pizza 55 | donut 56 | cake 57 | chair 58 | sofa 59 | pottedplant 60 | bed 61 | diningtable 62 | toilet 63 | tvmonitor 64 | laptop 65 | mouse 66 | remote 67 | keyboard 68 | cell phone 69 | microwave 70 | oven 71 | toaster 72 | sink 73 | refrigerator 74 | book 75 | clock 76 | vase 77 | scissors 78 | teddy bear 79 | hair drier 80 | toothbrush 81 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/cfg/darknet19_448.cfg: -------------------------------------------------------------------------------- 1 | [net] 2 | batch=128 3 | subdivisions=4 4 | height=448 5 | width=448 6 | max_crop=512 7 | channels=3 8 | momentum=0.9 9 | decay=0.0005 10 | 11 | learning_rate=0.001 12 | policy=poly 13 | power=4 14 | max_batches=100000 15 | 16 | angle=7 17 | hue = .1 18 | saturation=.75 19 | exposure=.75 20 | aspect=.75 21 | 22 | [convolutional] 23 | batch_normalize=1 24 | filters=32 25 | size=3 26 | stride=1 27 | pad=1 28 | activation=leaky 29 | 30 | [maxpool] 31 | size=2 32 | stride=2 33 | 34 | [convolutional] 35 | batch_normalize=1 36 | filters=64 37 | size=3 38 | stride=1 39 | pad=1 40 | activation=leaky 41 | 42 | [maxpool] 43 | size=2 44 | stride=2 45 | 46 | [convolutional] 47 | batch_normalize=1 48 | filters=128 49 | size=3 50 | stride=1 51 | pad=1 52 | activation=leaky 53 | 54 | [convolutional] 55 | batch_normalize=1 56 | filters=64 57 | size=1 58 | stride=1 59 | pad=1 60 | activation=leaky 61 | 62 | [convolutional] 63 | batch_normalize=1 64 | filters=128 65 | size=3 66 | stride=1 67 | pad=1 68 | activation=leaky 69 | 70 | [maxpool] 71 | size=2 72 | stride=2 73 | 74 | [convolutional] 75 | batch_normalize=1 76 | filters=256 77 | size=3 78 | stride=1 79 | pad=1 80 | activation=leaky 81 | 82 | [convolutional] 83 | batch_normalize=1 84 | filters=128 85 | size=1 86 | stride=1 87 | pad=1 88 | activation=leaky 89 | 90 | [convolutional] 91 | batch_normalize=1 92 | filters=256 93 | size=3 94 | stride=1 95 | pad=1 96 | activation=leaky 97 | 98 | [maxpool] 99 | size=2 100 | stride=2 101 | 102 | [convolutional] 103 | batch_normalize=1 104 | filters=512 105 | size=3 106 | stride=1 107 | pad=1 108 | activation=leaky 109 | 110 | [convolutional] 111 | batch_normalize=1 112 | filters=256 113 | size=1 114 | stride=1 115 | pad=1 116 | activation=leaky 117 | 118 | [convolutional] 119 | batch_normalize=1 120 | filters=512 121 | size=3 122 | stride=1 123 | pad=1 124 | activation=leaky 125 | 126 | [convolutional] 127 | batch_normalize=1 128 | filters=256 129 | size=1 130 | stride=1 131 | pad=1 132 | activation=leaky 133 | 134 | [convolutional] 135 | batch_normalize=1 136 | filters=512 137 | size=3 138 | stride=1 139 | pad=1 140 | activation=leaky 141 | 142 | [maxpool] 143 | size=2 144 | stride=2 145 | 146 | [convolutional] 147 | batch_normalize=1 148 | filters=1024 149 | size=3 150 | stride=1 151 | pad=1 152 | activation=leaky 153 | 154 | [convolutional] 155 | batch_normalize=1 156 | filters=512 157 | size=1 158 | stride=1 159 | pad=1 160 | activation=leaky 161 | 162 | [convolutional] 163 | batch_normalize=1 164 | filters=1024 165 | size=3 166 | stride=1 167 | pad=1 168 | activation=leaky 169 | 170 | [convolutional] 171 | batch_normalize=1 172 | filters=512 173 | size=1 174 | stride=1 175 | pad=1 176 | activation=leaky 177 | 178 | [convolutional] 179 | batch_normalize=1 180 | filters=1024 181 | size=3 182 | stride=1 183 | pad=1 184 | activation=leaky 185 | 186 | [convolutional] 187 | filters=1000 188 | size=1 189 | stride=1 190 | pad=1 191 | activation=linear 192 | 193 | [avgpool] 194 | 195 | [softmax] 196 | groups=1 197 | 198 | [cost] 199 | type=sse 200 | 201 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/cfg/tiny-yolo-voc.cfg: -------------------------------------------------------------------------------- 1 | [net] 2 | batch=64 3 | subdivisions=8 4 | width=416 5 | height=416 6 | channels=3 7 | momentum=0.9 8 | decay=0.0005 9 | angle=0 10 | saturation = 1.5 11 | exposure = 1.5 12 | hue=.1 13 | 14 | learning_rate=0.001 15 | max_batches = 40200 16 | policy=steps 17 | steps=-1,100,20000,30000 18 | scales=.1,10,.1,.1 19 | 20 | [convolutional] 21 | batch_normalize=1 22 | filters=16 23 | size=3 24 | stride=1 25 | pad=1 26 | activation=leaky 27 | 28 | [maxpool] 29 | size=2 30 | stride=2 31 | 32 | [convolutional] 33 | batch_normalize=1 34 | filters=32 35 | size=3 36 | stride=1 37 | pad=1 38 | activation=leaky 39 | 40 | [maxpool] 41 | size=2 42 | stride=2 43 | 44 | [convolutional] 45 | batch_normalize=1 46 | filters=64 47 | size=3 48 | stride=1 49 | pad=1 50 | activation=leaky 51 | 52 | [maxpool] 53 | size=2 54 | stride=2 55 | 56 | [convolutional] 57 | batch_normalize=1 58 | filters=128 59 | size=3 60 | stride=1 61 | pad=1 62 | activation=leaky 63 | 64 | [maxpool] 65 | size=2 66 | stride=2 67 | 68 | [convolutional] 69 | batch_normalize=1 70 | filters=256 71 | size=3 72 | stride=1 73 | pad=1 74 | activation=leaky 75 | 76 | [maxpool] 77 | size=2 78 | stride=2 79 | 80 | [convolutional] 81 | batch_normalize=1 82 | filters=512 83 | size=3 84 | stride=1 85 | pad=1 86 | activation=leaky 87 | 88 | [maxpool] 89 | size=2 90 | stride=1 91 | 92 | [convolutional] 93 | batch_normalize=1 94 | filters=1024 95 | size=3 96 | stride=1 97 | pad=1 98 | activation=leaky 99 | 100 | ########### 101 | 102 | [convolutional] 103 | batch_normalize=1 104 | size=3 105 | stride=1 106 | pad=1 107 | filters=1024 108 | activation=leaky 109 | 110 | [convolutional] 111 | size=1 112 | stride=1 113 | pad=1 114 | filters=125 115 | activation=linear 116 | 117 | [region] 118 | anchors = 1.08,1.19, 3.42,4.41, 6.63,11.38, 9.42,5.11, 16.62,10.52 119 | bias_match=1 120 | classes=20 121 | coords=4 122 | num=5 123 | softmax=1 124 | jitter=.2 125 | rescore=1 126 | 127 | object_scale=5 128 | noobject_scale=1 129 | class_scale=1 130 | coord_scale=1 131 | 132 | absolute=1 133 | thresh = .6 134 | random=1 135 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/cfg/tiny-yolo.cfg: -------------------------------------------------------------------------------- 1 | [net] 2 | # Training 3 | # batch=64 4 | # subdivisions=2 5 | # Testing 6 | batch=1 7 | subdivisions=1 8 | width=416 9 | height=416 10 | channels=3 11 | momentum=0.9 12 | decay=0.0005 13 | angle=0 14 | saturation = 1.5 15 | exposure = 1.5 16 | hue=.1 17 | 18 | learning_rate=0.001 19 | burn_in=1000 20 | max_batches = 500200 21 | policy=steps 22 | steps=400000,450000 23 | scales=.1,.1 24 | 25 | [convolutional] 26 | batch_normalize=1 27 | filters=16 28 | size=3 29 | stride=1 30 | pad=1 31 | activation=leaky 32 | 33 | [maxpool] 34 | size=2 35 | stride=2 36 | 37 | [convolutional] 38 | batch_normalize=1 39 | filters=32 40 | size=3 41 | stride=1 42 | pad=1 43 | activation=leaky 44 | 45 | [maxpool] 46 | size=2 47 | stride=2 48 | 49 | [convolutional] 50 | batch_normalize=1 51 | filters=64 52 | size=3 53 | stride=1 54 | pad=1 55 | activation=leaky 56 | 57 | [maxpool] 58 | size=2 59 | stride=2 60 | 61 | [convolutional] 62 | batch_normalize=1 63 | filters=128 64 | size=3 65 | stride=1 66 | pad=1 67 | activation=leaky 68 | 69 | [maxpool] 70 | size=2 71 | stride=2 72 | 73 | [convolutional] 74 | batch_normalize=1 75 | filters=256 76 | size=3 77 | stride=1 78 | pad=1 79 | activation=leaky 80 | 81 | [maxpool] 82 | size=2 83 | stride=2 84 | 85 | [convolutional] 86 | batch_normalize=1 87 | filters=512 88 | size=3 89 | stride=1 90 | pad=1 91 | activation=leaky 92 | 93 | [maxpool] 94 | size=2 95 | stride=1 96 | 97 | [convolutional] 98 | batch_normalize=1 99 | filters=1024 100 | size=3 101 | stride=1 102 | pad=1 103 | activation=leaky 104 | 105 | ########### 106 | 107 | [convolutional] 108 | batch_normalize=1 109 | size=3 110 | stride=1 111 | pad=1 112 | filters=512 113 | activation=leaky 114 | 115 | [convolutional] 116 | size=1 117 | stride=1 118 | pad=1 119 | filters=425 120 | activation=linear 121 | 122 | [region] 123 | anchors = 0.57273, 0.677385, 1.87446, 2.06253, 3.33843, 5.47434, 7.88282, 3.52778, 9.77052, 9.16828 124 | bias_match=1 125 | classes=80 126 | coords=4 127 | num=5 128 | softmax=1 129 | jitter=.2 130 | rescore=0 131 | 132 | object_scale=5 133 | noobject_scale=1 134 | class_scale=1 135 | coord_scale=1 136 | 137 | absolute=1 138 | thresh = .6 139 | random=1 140 | 141 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/cfg/voc.data: -------------------------------------------------------------------------------- 1 | train = data/voc_train.txt 2 | valid = data/2007_test.txt 3 | names = data/voc.names 4 | backup = backup 5 | gpus = 3 6 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/cfg/voc.names: -------------------------------------------------------------------------------- 1 | aeroplane 2 | bicycle 3 | bird 4 | boat 5 | bottle 6 | bus 7 | car 8 | cat 9 | chair 10 | cow 11 | diningtable 12 | dog 13 | horse 14 | motorbike 15 | person 16 | pottedplant 17 | sheep 18 | sofa 19 | train 20 | tvmonitor 21 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/cfg/voc_gaotie.data: -------------------------------------------------------------------------------- 1 | train = data/gaotie_trainval.txt 2 | valid = data/gaotie_test.txt 3 | names = data/voc.names 4 | backup = backup 5 | gpus = 3 -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/cfg/yolo-voc.cfg: -------------------------------------------------------------------------------- 1 | [net] 2 | # Testing 3 | batch=64 4 | subdivisions=8 5 | # Training 6 | # batch=64 7 | # subdivisions=8 8 | height=416 9 | width=416 10 | channels=3 11 | momentum=0.9 12 | decay=0.0005 13 | angle=0 14 | saturation = 1.5 15 | exposure = 1.5 16 | hue=.1 17 | 18 | learning_rate=0.001 19 | burn_in=1000 20 | max_batches = 80200 21 | policy=steps 22 | steps=-1,500,40000,60000 23 | scales=0.1,10,.1,.1 24 | 25 | [convolutional] 26 | batch_normalize=1 27 | filters=32 28 | size=3 29 | stride=1 30 | pad=1 31 | activation=leaky 32 | 33 | [maxpool] 34 | size=2 35 | stride=2 36 | 37 | [convolutional] 38 | batch_normalize=1 39 | filters=64 40 | size=3 41 | stride=1 42 | pad=1 43 | activation=leaky 44 | 45 | [maxpool] 46 | size=2 47 | stride=2 48 | 49 | [convolutional] 50 | batch_normalize=1 51 | filters=128 52 | size=3 53 | stride=1 54 | pad=1 55 | activation=leaky 56 | 57 | [convolutional] 58 | batch_normalize=1 59 | filters=64 60 | size=1 61 | stride=1 62 | pad=1 63 | activation=leaky 64 | 65 | [convolutional] 66 | batch_normalize=1 67 | filters=128 68 | size=3 69 | stride=1 70 | pad=1 71 | activation=leaky 72 | 73 | [maxpool] 74 | size=2 75 | stride=2 76 | 77 | [convolutional] 78 | batch_normalize=1 79 | filters=256 80 | size=3 81 | stride=1 82 | pad=1 83 | activation=leaky 84 | 85 | [convolutional] 86 | batch_normalize=1 87 | filters=128 88 | size=1 89 | stride=1 90 | pad=1 91 | activation=leaky 92 | 93 | [convolutional] 94 | batch_normalize=1 95 | filters=256 96 | size=3 97 | stride=1 98 | pad=1 99 | activation=leaky 100 | 101 | [maxpool] 102 | size=2 103 | stride=2 104 | 105 | [convolutional] 106 | batch_normalize=1 107 | filters=512 108 | size=3 109 | stride=1 110 | pad=1 111 | activation=leaky 112 | 113 | [convolutional] 114 | batch_normalize=1 115 | filters=256 116 | size=1 117 | stride=1 118 | pad=1 119 | activation=leaky 120 | 121 | [convolutional] 122 | batch_normalize=1 123 | filters=512 124 | size=3 125 | stride=1 126 | pad=1 127 | activation=leaky 128 | 129 | [convolutional] 130 | batch_normalize=1 131 | filters=256 132 | size=1 133 | stride=1 134 | pad=1 135 | activation=leaky 136 | 137 | [convolutional] 138 | batch_normalize=1 139 | filters=512 140 | size=3 141 | stride=1 142 | pad=1 143 | activation=leaky 144 | 145 | [maxpool] 146 | size=2 147 | stride=2 148 | 149 | [convolutional] 150 | batch_normalize=1 151 | filters=1024 152 | size=3 153 | stride=1 154 | pad=1 155 | activation=leaky 156 | 157 | [convolutional] 158 | batch_normalize=1 159 | filters=512 160 | size=1 161 | stride=1 162 | pad=1 163 | activation=leaky 164 | 165 | [convolutional] 166 | batch_normalize=1 167 | filters=1024 168 | size=3 169 | stride=1 170 | pad=1 171 | activation=leaky 172 | 173 | [convolutional] 174 | batch_normalize=1 175 | filters=512 176 | size=1 177 | stride=1 178 | pad=1 179 | activation=leaky 180 | 181 | [convolutional] 182 | batch_normalize=1 183 | filters=1024 184 | size=3 185 | stride=1 186 | pad=1 187 | activation=leaky 188 | 189 | 190 | ####### 191 | 192 | [convolutional] 193 | batch_normalize=1 194 | size=3 195 | stride=1 196 | pad=1 197 | filters=1024 198 | activation=leaky 199 | 200 | [convolutional] 201 | batch_normalize=1 202 | size=3 203 | stride=1 204 | pad=1 205 | filters=1024 206 | activation=leaky 207 | 208 | [route] 209 | layers=-9 210 | 211 | [convolutional] 212 | batch_normalize=1 213 | size=1 214 | stride=1 215 | pad=1 216 | filters=64 217 | activation=leaky 218 | 219 | [reorg] 220 | stride=2 221 | 222 | [route] 223 | layers=-1,-4 224 | 225 | [convolutional] 226 | batch_normalize=1 227 | size=3 228 | stride=1 229 | pad=1 230 | filters=1024 231 | activation=leaky 232 | 233 | [convolutional] 234 | size=1 235 | stride=1 236 | pad=1 237 | filters=125 238 | activation=linear 239 | 240 | 241 | [region] 242 | anchors = 1.3221, 1.73145, 3.19275, 4.00944, 5.05587, 8.09892, 9.47112, 4.84053, 11.2364, 10.0071 243 | bias_match=1 244 | classes=20 245 | coords=4 246 | num=5 247 | softmax=1 248 | jitter=.3 249 | rescore=1 250 | 251 | object_scale=5 252 | noobject_scale=1 253 | class_scale=1 254 | coord_scale=1 255 | 256 | absolute=1 257 | thresh = .6 258 | random=1 259 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/cfg/yolo.cfg: -------------------------------------------------------------------------------- 1 | [net] 2 | # Testing 3 | batch=1 4 | subdivisions=1 5 | # Training 6 | # batch=64 7 | # subdivisions=8 8 | width=416 9 | height=416 10 | channels=3 11 | momentum=0.9 12 | decay=0.0005 13 | angle=0 14 | saturation = 1.5 15 | exposure = 1.5 16 | hue=.1 17 | 18 | learning_rate=0.001 19 | burn_in=1000 20 | max_batches = 500200 21 | policy=steps 22 | steps=400000,450000 23 | scales=.1,.1 24 | 25 | [convolutional] 26 | batch_normalize=1 27 | filters=32 28 | size=3 29 | stride=1 30 | pad=1 31 | activation=leaky 32 | 33 | [maxpool] 34 | size=2 35 | stride=2 36 | 37 | [convolutional] 38 | batch_normalize=1 39 | filters=64 40 | size=3 41 | stride=1 42 | pad=1 43 | activation=leaky 44 | 45 | [maxpool] 46 | size=2 47 | stride=2 48 | 49 | [convolutional] 50 | batch_normalize=1 51 | filters=128 52 | size=3 53 | stride=1 54 | pad=1 55 | activation=leaky 56 | 57 | [convolutional] 58 | batch_normalize=1 59 | filters=64 60 | size=1 61 | stride=1 62 | pad=1 63 | activation=leaky 64 | 65 | [convolutional] 66 | batch_normalize=1 67 | filters=128 68 | size=3 69 | stride=1 70 | pad=1 71 | activation=leaky 72 | 73 | [maxpool] 74 | size=2 75 | stride=2 76 | 77 | [convolutional] 78 | batch_normalize=1 79 | filters=256 80 | size=3 81 | stride=1 82 | pad=1 83 | activation=leaky 84 | 85 | [convolutional] 86 | batch_normalize=1 87 | filters=128 88 | size=1 89 | stride=1 90 | pad=1 91 | activation=leaky 92 | 93 | [convolutional] 94 | batch_normalize=1 95 | filters=256 96 | size=3 97 | stride=1 98 | pad=1 99 | activation=leaky 100 | 101 | [maxpool] 102 | size=2 103 | stride=2 104 | 105 | [convolutional] 106 | batch_normalize=1 107 | filters=512 108 | size=3 109 | stride=1 110 | pad=1 111 | activation=leaky 112 | 113 | [convolutional] 114 | batch_normalize=1 115 | filters=256 116 | size=1 117 | stride=1 118 | pad=1 119 | activation=leaky 120 | 121 | [convolutional] 122 | batch_normalize=1 123 | filters=512 124 | size=3 125 | stride=1 126 | pad=1 127 | activation=leaky 128 | 129 | [convolutional] 130 | batch_normalize=1 131 | filters=256 132 | size=1 133 | stride=1 134 | pad=1 135 | activation=leaky 136 | 137 | [convolutional] 138 | batch_normalize=1 139 | filters=512 140 | size=3 141 | stride=1 142 | pad=1 143 | activation=leaky 144 | 145 | [maxpool] 146 | size=2 147 | stride=2 148 | 149 | [convolutional] 150 | batch_normalize=1 151 | filters=1024 152 | size=3 153 | stride=1 154 | pad=1 155 | activation=leaky 156 | 157 | [convolutional] 158 | batch_normalize=1 159 | filters=512 160 | size=1 161 | stride=1 162 | pad=1 163 | activation=leaky 164 | 165 | [convolutional] 166 | batch_normalize=1 167 | filters=1024 168 | size=3 169 | stride=1 170 | pad=1 171 | activation=leaky 172 | 173 | [convolutional] 174 | batch_normalize=1 175 | filters=512 176 | size=1 177 | stride=1 178 | pad=1 179 | activation=leaky 180 | 181 | [convolutional] 182 | batch_normalize=1 183 | filters=1024 184 | size=3 185 | stride=1 186 | pad=1 187 | activation=leaky 188 | 189 | 190 | ####### 191 | 192 | [convolutional] 193 | batch_normalize=1 194 | size=3 195 | stride=1 196 | pad=1 197 | filters=1024 198 | activation=leaky 199 | 200 | [convolutional] 201 | batch_normalize=1 202 | size=3 203 | stride=1 204 | pad=1 205 | filters=1024 206 | activation=leaky 207 | 208 | [route] 209 | layers=-9 210 | 211 | [convolutional] 212 | batch_normalize=1 213 | size=1 214 | stride=1 215 | pad=1 216 | filters=64 217 | activation=leaky 218 | 219 | [reorg] 220 | stride=2 221 | 222 | [route] 223 | layers=-1,-4 224 | 225 | [convolutional] 226 | batch_normalize=1 227 | size=3 228 | stride=1 229 | pad=1 230 | filters=1024 231 | activation=leaky 232 | 233 | [convolutional] 234 | size=1 235 | stride=1 236 | pad=1 237 | filters=425 238 | activation=linear 239 | 240 | 241 | [region] 242 | anchors = 0.57273, 0.677385, 1.87446, 2.06253, 3.33843, 5.47434, 7.88282, 3.52778, 9.77052, 9.16828 243 | bias_match=1 244 | classes=80 245 | coords=4 246 | num=5 247 | softmax=1 248 | jitter=.3 249 | rescore=1 250 | 251 | object_scale=5 252 | noobject_scale=1 253 | class_scale=1 254 | coord_scale=1 255 | 256 | absolute=1 257 | thresh = .6 258 | random=1 259 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/cfg/yolov3-tiny.cfg: -------------------------------------------------------------------------------- 1 | [net] 2 | # Testing 3 | batch=1 4 | subdivisions=1 5 | # Training 6 | # batch=64 7 | # subdivisions=2 8 | width=416 9 | height=416 10 | channels=3 11 | momentum=0.9 12 | decay=0.0005 13 | angle=0 14 | saturation = 1.5 15 | exposure = 1.5 16 | hue=.1 17 | 18 | learning_rate=0.001 19 | burn_in=1000 20 | max_batches = 500200 21 | policy=steps 22 | steps=400000,450000 23 | scales=.1,.1 24 | 25 | [convolutional] 26 | batch_normalize=1 27 | filters=16 28 | size=3 29 | stride=1 30 | pad=1 31 | activation=leaky 32 | 33 | [maxpool] 34 | size=2 35 | stride=2 36 | 37 | [convolutional] 38 | batch_normalize=1 39 | filters=32 40 | size=3 41 | stride=1 42 | pad=1 43 | activation=leaky 44 | 45 | [maxpool] 46 | size=2 47 | stride=2 48 | 49 | [convolutional] 50 | batch_normalize=1 51 | filters=64 52 | size=3 53 | stride=1 54 | pad=1 55 | activation=leaky 56 | 57 | [maxpool] 58 | size=2 59 | stride=2 60 | 61 | [convolutional] 62 | batch_normalize=1 63 | filters=128 64 | size=3 65 | stride=1 66 | pad=1 67 | activation=leaky 68 | 69 | [maxpool] 70 | size=2 71 | stride=2 72 | 73 | [convolutional] 74 | batch_normalize=1 75 | filters=256 76 | size=3 77 | stride=1 78 | pad=1 79 | activation=leaky 80 | 81 | [maxpool] 82 | size=2 83 | stride=2 84 | 85 | [convolutional] 86 | batch_normalize=1 87 | filters=512 88 | size=3 89 | stride=1 90 | pad=1 91 | activation=leaky 92 | 93 | [maxpool] 94 | size=2 95 | stride=1 96 | 97 | [convolutional] 98 | batch_normalize=1 99 | filters=1024 100 | size=3 101 | stride=1 102 | pad=1 103 | activation=leaky 104 | 105 | ########### 106 | 107 | [convolutional] 108 | batch_normalize=1 109 | filters=256 110 | size=1 111 | stride=1 112 | pad=1 113 | activation=leaky 114 | 115 | [convolutional] 116 | batch_normalize=1 117 | filters=512 118 | size=3 119 | stride=1 120 | pad=1 121 | activation=leaky 122 | 123 | [convolutional] 124 | size=1 125 | stride=1 126 | pad=1 127 | filters=255 128 | activation=linear 129 | 130 | 131 | 132 | [yolo] 133 | mask = 3,4,5 134 | anchors = 10,14, 23,27, 37,58, 81,82, 135,169, 344,319 135 | classes=80 136 | num=6 137 | jitter=.3 138 | ignore_thresh = .7 139 | truth_thresh = 1 140 | random=1 141 | 142 | [route] 143 | layers = -4 144 | 145 | [convolutional] 146 | batch_normalize=1 147 | filters=128 148 | size=1 149 | stride=1 150 | pad=1 151 | activation=leaky 152 | 153 | [upsample] 154 | stride=2 155 | 156 | [route] 157 | layers = -1, 8 158 | 159 | [convolutional] 160 | batch_normalize=1 161 | filters=256 162 | size=3 163 | stride=1 164 | pad=1 165 | activation=leaky 166 | 167 | [convolutional] 168 | size=1 169 | stride=1 170 | pad=1 171 | filters=255 172 | activation=linear 173 | 174 | [yolo] 175 | mask = 0,1,2 176 | anchors = 10,14, 23,27, 37,58, 81,82, 135,169, 344,319 177 | classes=80 178 | num=6 179 | jitter=.3 180 | ignore_thresh = .7 181 | truth_thresh = 1 182 | random=1 183 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/demo/004545.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/deep_sort/detector/YOLOv3/demo/004545.jpg -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/demo/results/004545.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/deep_sort/detector/YOLOv3/demo/results/004545.jpg -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/detect.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import time 3 | from PIL import Image, ImageDraw 4 | #from models.tiny_yolo import TinyYoloNet 5 | from yolo_utils import * 6 | from darknet import Darknet 7 | 8 | import cv2 9 | 10 | namesfile=None 11 | def detect(cfgfile, weightfile, imgfolder): 12 | m = Darknet(cfgfile) 13 | 14 | #m.print_network() 15 | m.load_weights(weightfile) 16 | print('Loading weights from %s... Done!' % (weightfile)) 17 | 18 | # if m.num_classes == 20: 19 | # namesfile = 'data/voc.names' 20 | # elif m.num_classes == 80: 21 | # namesfile = 'data/coco.names' 22 | # else: 23 | # namesfile = 'data/names' 24 | 25 | use_cuda = True 26 | if use_cuda: 27 | m.cuda() 28 | 29 | imgfiles = [x for x in os.listdir(imgfolder) if x[-4:] == '.jpg'] 30 | imgfiles.sort() 31 | for imgname in imgfiles: 32 | imgfile = os.path.join(imgfolder,imgname) 33 | 34 | img = Image.open(imgfile).convert('RGB') 35 | sized = img.resize((m.width, m.height)) 36 | 37 | #for i in range(2): 38 | start = time.time() 39 | boxes = do_detect(m, sized, 0.5, 0.4, use_cuda) 40 | finish = time.time() 41 | #if i == 1: 42 | print('%s: Predicted in %f seconds.' % (imgfile, (finish-start))) 43 | 44 | class_names = load_class_names(namesfile) 45 | img = plot_boxes(img, boxes, 'result/{}'.format(os.path.basename(imgfile)), class_names) 46 | img = np.array(img) 47 | cv2.imshow('{}'.format(os.path.basename(imgfolder)), img) 48 | cv2.resizeWindow('{}'.format(os.path.basename(imgfolder)), 1000,800) 49 | cv2.waitKey(1000) 50 | 51 | def detect_cv2(cfgfile, weightfile, imgfile): 52 | import cv2 53 | m = Darknet(cfgfile) 54 | 55 | m.print_network() 56 | m.load_weights(weightfile) 57 | print('Loading weights from %s... Done!' % (weightfile)) 58 | 59 | if m.num_classes == 20: 60 | namesfile = 'data/voc.names' 61 | elif m.num_classes == 80: 62 | namesfile = 'data/coco.names' 63 | else: 64 | namesfile = 'data/names' 65 | 66 | use_cuda = True 67 | if use_cuda: 68 | m.cuda() 69 | 70 | img = cv2.imread(imgfile) 71 | sized = cv2.resize(img, (m.width, m.height)) 72 | sized = cv2.cvtColor(sized, cv2.COLOR_BGR2RGB) 73 | 74 | for i in range(2): 75 | start = time.time() 76 | boxes = do_detect(m, sized, 0.5, 0.4, use_cuda) 77 | finish = time.time() 78 | if i == 1: 79 | print('%s: Predicted in %f seconds.' % (imgfile, (finish-start))) 80 | 81 | class_names = load_class_names(namesfile) 82 | plot_boxes_cv2(img, boxes, savename='predictions.jpg', class_names=class_names) 83 | 84 | def detect_skimage(cfgfile, weightfile, imgfile): 85 | from skimage import io 86 | from skimage.transform import resize 87 | m = Darknet(cfgfile) 88 | 89 | m.print_network() 90 | m.load_weights(weightfile) 91 | print('Loading weights from %s... Done!' % (weightfile)) 92 | 93 | if m.num_classes == 20: 94 | namesfile = 'data/voc.names' 95 | elif m.num_classes == 80: 96 | namesfile = 'data/coco.names' 97 | else: 98 | namesfile = 'data/names' 99 | 100 | use_cuda = True 101 | if use_cuda: 102 | m.cuda() 103 | 104 | img = io.imread(imgfile) 105 | sized = resize(img, (m.width, m.height)) * 255 106 | 107 | for i in range(2): 108 | start = time.time() 109 | boxes = do_detect(m, sized, 0.5, 0.4, use_cuda) 110 | finish = time.time() 111 | if i == 1: 112 | print('%s: Predicted in %f seconds.' % (imgfile, (finish-start))) 113 | 114 | class_names = load_class_names(namesfile) 115 | plot_boxes_cv2(img, boxes, savename='predictions.jpg', class_names=class_names) 116 | 117 | if __name__ == '__main__': 118 | if len(sys.argv) == 5: 119 | cfgfile = sys.argv[1] 120 | weightfile = sys.argv[2] 121 | imgfolder = sys.argv[3] 122 | cv2.namedWindow('{}'.format(os.path.basename(imgfolder)), cv2.WINDOW_NORMAL ) 123 | cv2.resizeWindow('{}'.format(os.path.basename(imgfolder)), 1000,800) 124 | globals()["namesfile"] = sys.argv[4] 125 | detect(cfgfile, weightfile, imgfolder) 126 | #detect_cv2(cfgfile, weightfile, imgfile) 127 | #detect_skimage(cfgfile, weightfile, imgfile) 128 | else: 129 | print('Usage: ') 130 | print(' python detect.py cfgfile weightfile imgfolder names') 131 | #detect('cfg/tiny-yolo-voc.cfg', 'tiny-yolo-voc.weights', 'data/person.jpg', version=1) 132 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/detector.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import logging 3 | import numpy as np 4 | import cv2 5 | 6 | from .darknet import Darknet 7 | from .yolo_utils import get_all_boxes, nms, post_process, xywh_to_xyxy, xyxy_to_xywh 8 | from .nms import boxes_nms 9 | 10 | 11 | class YOLOv3(object): 12 | def __init__(self, cfgfile, weightfile, namesfile, score_thresh=0.7, conf_thresh=0.01, nms_thresh=0.45, 13 | is_xywh=False, use_cuda=True): 14 | # net definition 15 | self.net = Darknet(cfgfile) 16 | self.net.load_weights(weightfile) 17 | logger = logging.getLogger("root.detector") 18 | logger.info('Loading weights from %s... Done!' % (weightfile)) 19 | self.device = "cuda" if use_cuda else "cpu" 20 | self.net.eval() 21 | self.net.to(self.device) 22 | 23 | # constants 24 | self.size = self.net.width, self.net.height 25 | self.score_thresh = score_thresh 26 | self.conf_thresh = conf_thresh 27 | self.nms_thresh = nms_thresh 28 | self.use_cuda = use_cuda 29 | self.is_xywh = is_xywh 30 | self.num_classes = self.net.num_classes 31 | self.class_names = self.load_class_names(namesfile) 32 | 33 | def __call__(self, ori_img): 34 | # img to tensor 35 | assert isinstance(ori_img, np.ndarray), "input must be a numpy array!" 36 | img = ori_img.astype(np.float) / 255. 37 | 38 | img = cv2.resize(img, self.size) 39 | img = torch.from_numpy(img).float().permute(2, 0, 1).unsqueeze(0) 40 | 41 | # forward 42 | with torch.no_grad(): 43 | img = img.to(self.device) 44 | out_boxes = self.net(img) 45 | boxes = get_all_boxes(out_boxes, self.conf_thresh, self.num_classes, 46 | use_cuda=self.use_cuda) # batch size is 1 47 | # boxes = nms(boxes, self.nms_thresh) 48 | 49 | boxes = post_process(boxes, self.net.num_classes, self.conf_thresh, self.nms_thresh)[0].cpu() 50 | boxes = boxes[boxes[:, -2] > self.score_thresh, :] # bbox xmin ymin xmax ymax 51 | 52 | if len(boxes) == 0: 53 | bbox = torch.FloatTensor([]).reshape([0, 4]) 54 | cls_conf = torch.FloatTensor([]) 55 | cls_ids = torch.LongTensor([]) 56 | else: 57 | height, width = ori_img.shape[:2] 58 | bbox = boxes[:, :4] 59 | if self.is_xywh: 60 | # bbox x y w h 61 | bbox = xyxy_to_xywh(bbox) 62 | 63 | bbox *= torch.FloatTensor([[width, height, width, height]]) 64 | cls_conf = boxes[:, 5] 65 | cls_ids = boxes[:, 6].long() 66 | return bbox.numpy(), cls_conf.numpy(), cls_ids.numpy() 67 | 68 | def load_class_names(self, namesfile): 69 | with open(namesfile, 'r', encoding='utf8') as fp: 70 | class_names = [line.strip() for line in fp.readlines()] 71 | return class_names 72 | 73 | 74 | def demo(): 75 | import os 76 | from vizer.draw import draw_boxes 77 | 78 | yolo = YOLOv3("cfg/yolo_v3.cfg", "weight/yolov3.weights", "cfg/coco.names") 79 | print("yolo.size =", yolo.size) 80 | root = "./demo" 81 | resdir = os.path.join(root, "results") 82 | os.makedirs(resdir, exist_ok=True) 83 | files = [os.path.join(root, file) for file in os.listdir(root) if file.endswith('.jpg')] 84 | files.sort() 85 | for filename in files: 86 | img = cv2.imread(filename) 87 | img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) 88 | bbox, cls_conf, cls_ids = yolo(img) 89 | 90 | if bbox is not None: 91 | img = draw_boxes(img, bbox, cls_ids, cls_conf, class_name_map=yolo.class_names) 92 | # save results 93 | cv2.imwrite(os.path.join(resdir, os.path.basename(filename)), img[:, :, (2, 1, 0)]) 94 | # imshow 95 | # cv2.namedWindow("yolo", cv2.WINDOW_NORMAL) 96 | # cv2.resizeWindow("yolo", 600,600) 97 | # cv2.imshow("yolo",res[:,:,(2,1,0)]) 98 | # cv2.waitKey(0) 99 | 100 | 101 | if __name__ == "__main__": 102 | demo() 103 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/nms/__init__.py: -------------------------------------------------------------------------------- 1 | from .nms import boxes_nms -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/nms/build.sh: -------------------------------------------------------------------------------- 1 | cd ext 2 | 3 | python build.py build_ext develop 4 | 5 | cd .. 6 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/nms/ext/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/deep_sort/detector/YOLOv3/nms/ext/__init__.py -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/nms/ext/build.py: -------------------------------------------------------------------------------- 1 | import glob 2 | import os 3 | 4 | import torch 5 | from setuptools import setup 6 | from torch.utils.cpp_extension import CUDA_HOME 7 | from torch.utils.cpp_extension import CppExtension 8 | from torch.utils.cpp_extension import CUDAExtension 9 | 10 | requirements = ["torch"] 11 | 12 | 13 | def get_extensions(): 14 | extensions_dir = os.path.dirname(os.path.abspath(__file__)) 15 | 16 | main_file = glob.glob(os.path.join(extensions_dir, "*.cpp")) 17 | source_cpu = glob.glob(os.path.join(extensions_dir, "cpu", "*.cpp")) 18 | source_cuda = glob.glob(os.path.join(extensions_dir, "cuda", "*.cu")) 19 | 20 | sources = main_file + source_cpu 21 | extension = CppExtension 22 | 23 | extra_compile_args = {"cxx": []} 24 | define_macros = [] 25 | 26 | if torch.cuda.is_available() and CUDA_HOME is not None: 27 | extension = CUDAExtension 28 | sources += source_cuda 29 | define_macros += [("WITH_CUDA", None)] 30 | extra_compile_args["nvcc"] = [ 31 | "-DCUDA_HAS_FP16=1", 32 | "-D__CUDA_NO_HALF_OPERATORS__", 33 | "-D__CUDA_NO_HALF_CONVERSIONS__", 34 | "-D__CUDA_NO_HALF2_OPERATORS__", 35 | ] 36 | 37 | sources = [os.path.join(extensions_dir, s) for s in sources] 38 | 39 | include_dirs = [extensions_dir] 40 | 41 | ext_modules = [ 42 | extension( 43 | "torch_extension", 44 | sources, 45 | include_dirs=include_dirs, 46 | define_macros=define_macros, 47 | extra_compile_args=extra_compile_args, 48 | ) 49 | ] 50 | 51 | return ext_modules 52 | 53 | 54 | setup( 55 | name="torch_extension", 56 | version="0.1", 57 | ext_modules=get_extensions(), 58 | cmdclass={"build_ext": torch.utils.cpp_extension.BuildExtension}) 59 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/nms/ext/cpu/nms_cpu.cpp: -------------------------------------------------------------------------------- 1 | // Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | #include "cpu/vision.h" 3 | 4 | 5 | template 6 | at::Tensor nms_cpu_kernel(const at::Tensor& dets, 7 | const at::Tensor& scores, 8 | const float threshold) { 9 | AT_ASSERTM(!dets.type().is_cuda(), "dets must be a CPU tensor"); 10 | AT_ASSERTM(!scores.type().is_cuda(), "scores must be a CPU tensor"); 11 | AT_ASSERTM(dets.type() == scores.type(), "dets should have the same type as scores"); 12 | 13 | if (dets.numel() == 0) { 14 | return at::empty({0}, dets.options().dtype(at::kLong).device(at::kCPU)); 15 | } 16 | 17 | auto x1_t = dets.select(1, 0).contiguous(); 18 | auto y1_t = dets.select(1, 1).contiguous(); 19 | auto x2_t = dets.select(1, 2).contiguous(); 20 | auto y2_t = dets.select(1, 3).contiguous(); 21 | 22 | at::Tensor areas_t = (x2_t - x1_t) * (y2_t - y1_t); 23 | 24 | auto order_t = std::get<1>(scores.sort(0, /* descending=*/true)); 25 | 26 | auto ndets = dets.size(0); 27 | at::Tensor suppressed_t = at::zeros({ndets}, dets.options().dtype(at::kByte).device(at::kCPU)); 28 | 29 | auto suppressed = suppressed_t.data(); 30 | auto order = order_t.data(); 31 | auto x1 = x1_t.data(); 32 | auto y1 = y1_t.data(); 33 | auto x2 = x2_t.data(); 34 | auto y2 = y2_t.data(); 35 | auto areas = areas_t.data(); 36 | 37 | for (int64_t _i = 0; _i < ndets; _i++) { 38 | auto i = order[_i]; 39 | if (suppressed[i] == 1) 40 | continue; 41 | auto ix1 = x1[i]; 42 | auto iy1 = y1[i]; 43 | auto ix2 = x2[i]; 44 | auto iy2 = y2[i]; 45 | auto iarea = areas[i]; 46 | 47 | for (int64_t _j = _i + 1; _j < ndets; _j++) { 48 | auto j = order[_j]; 49 | if (suppressed[j] == 1) 50 | continue; 51 | auto xx1 = std::max(ix1, x1[j]); 52 | auto yy1 = std::max(iy1, y1[j]); 53 | auto xx2 = std::min(ix2, x2[j]); 54 | auto yy2 = std::min(iy2, y2[j]); 55 | 56 | auto w = std::max(static_cast(0), xx2 - xx1); 57 | auto h = std::max(static_cast(0), yy2 - yy1); 58 | auto inter = w * h; 59 | auto ovr = inter / (iarea + areas[j] - inter); 60 | if (ovr >= threshold) 61 | suppressed[j] = 1; 62 | } 63 | } 64 | return at::nonzero(suppressed_t == 0).squeeze(1); 65 | } 66 | 67 | at::Tensor nms_cpu(const at::Tensor& dets, 68 | const at::Tensor& scores, 69 | const float threshold) { 70 | at::Tensor result; 71 | AT_DISPATCH_FLOATING_TYPES(dets.type(), "nms", [&] { 72 | result = nms_cpu_kernel(dets, scores, threshold); 73 | }); 74 | return result; 75 | } -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/nms/ext/cpu/vision.h: -------------------------------------------------------------------------------- 1 | // Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | #pragma once 3 | #include 4 | 5 | at::Tensor nms_cpu(const at::Tensor& dets, 6 | const at::Tensor& scores, 7 | const float threshold); 8 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/nms/ext/cuda/nms.cu: -------------------------------------------------------------------------------- 1 | // Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | #include 3 | #include 4 | 5 | #include 6 | #include 7 | 8 | #include 9 | #include 10 | 11 | int const threadsPerBlock = sizeof(unsigned long long) * 8; 12 | 13 | __device__ inline float devIoU(float const * const a, float const * const b) { 14 | float left = max(a[0], b[0]), right = min(a[2], b[2]); 15 | float top = max(a[1], b[1]), bottom = min(a[3], b[3]); 16 | float width = max(right - left, 0.f), height = max(bottom - top, 0.f); 17 | float interS = width * height; 18 | float Sa = (a[2] - a[0]) * (a[3] - a[1]); 19 | float Sb = (b[2] - b[0]) * (b[3] - b[1]); 20 | return interS / (Sa + Sb - interS); 21 | } 22 | 23 | __global__ void nms_kernel(const int n_boxes, const float nms_overlap_thresh, 24 | const float *dev_boxes, unsigned long long *dev_mask) { 25 | const int row_start = blockIdx.y; 26 | const int col_start = blockIdx.x; 27 | 28 | // if (row_start > col_start) return; 29 | 30 | const int row_size = 31 | min(n_boxes - row_start * threadsPerBlock, threadsPerBlock); 32 | const int col_size = 33 | min(n_boxes - col_start * threadsPerBlock, threadsPerBlock); 34 | 35 | __shared__ float block_boxes[threadsPerBlock * 5]; 36 | if (threadIdx.x < col_size) { 37 | block_boxes[threadIdx.x * 5 + 0] = 38 | dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 0]; 39 | block_boxes[threadIdx.x * 5 + 1] = 40 | dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 1]; 41 | block_boxes[threadIdx.x * 5 + 2] = 42 | dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 2]; 43 | block_boxes[threadIdx.x * 5 + 3] = 44 | dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 3]; 45 | block_boxes[threadIdx.x * 5 + 4] = 46 | dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 4]; 47 | } 48 | __syncthreads(); 49 | 50 | if (threadIdx.x < row_size) { 51 | const int cur_box_idx = threadsPerBlock * row_start + threadIdx.x; 52 | const float *cur_box = dev_boxes + cur_box_idx * 5; 53 | int i = 0; 54 | unsigned long long t = 0; 55 | int start = 0; 56 | if (row_start == col_start) { 57 | start = threadIdx.x + 1; 58 | } 59 | for (i = start; i < col_size; i++) { 60 | if (devIoU(cur_box, block_boxes + i * 5) > nms_overlap_thresh) { 61 | t |= 1ULL << i; 62 | } 63 | } 64 | const int col_blocks = THCCeilDiv(n_boxes, threadsPerBlock); 65 | dev_mask[cur_box_idx * col_blocks + col_start] = t; 66 | } 67 | } 68 | 69 | // boxes is a N x 5 tensor 70 | at::Tensor nms_cuda(const at::Tensor boxes, float nms_overlap_thresh) { 71 | using scalar_t = float; 72 | AT_ASSERTM(boxes.type().is_cuda(), "boxes must be a CUDA tensor"); 73 | auto scores = boxes.select(1, 4); 74 | auto order_t = std::get<1>(scores.sort(0, /* descending=*/true)); 75 | auto boxes_sorted = boxes.index_select(0, order_t); 76 | 77 | int boxes_num = boxes.size(0); 78 | 79 | const int col_blocks = THCCeilDiv(boxes_num, threadsPerBlock); 80 | 81 | scalar_t* boxes_dev = boxes_sorted.data(); 82 | 83 | THCState *state = at::globalContext().lazyInitCUDA(); // TODO replace with getTHCState 84 | 85 | unsigned long long* mask_dev = NULL; 86 | //THCudaCheck(THCudaMalloc(state, (void**) &mask_dev, 87 | // boxes_num * col_blocks * sizeof(unsigned long long))); 88 | 89 | mask_dev = (unsigned long long*) THCudaMalloc(state, boxes_num * col_blocks * sizeof(unsigned long long)); 90 | 91 | dim3 blocks(THCCeilDiv(boxes_num, threadsPerBlock), 92 | THCCeilDiv(boxes_num, threadsPerBlock)); 93 | dim3 threads(threadsPerBlock); 94 | nms_kernel<<>>(boxes_num, 95 | nms_overlap_thresh, 96 | boxes_dev, 97 | mask_dev); 98 | 99 | std::vector mask_host(boxes_num * col_blocks); 100 | THCudaCheck(cudaMemcpy(&mask_host[0], 101 | mask_dev, 102 | sizeof(unsigned long long) * boxes_num * col_blocks, 103 | cudaMemcpyDeviceToHost)); 104 | 105 | std::vector remv(col_blocks); 106 | memset(&remv[0], 0, sizeof(unsigned long long) * col_blocks); 107 | 108 | at::Tensor keep = at::empty({boxes_num}, boxes.options().dtype(at::kLong).device(at::kCPU)); 109 | int64_t* keep_out = keep.data(); 110 | 111 | int num_to_keep = 0; 112 | for (int i = 0; i < boxes_num; i++) { 113 | int nblock = i / threadsPerBlock; 114 | int inblock = i % threadsPerBlock; 115 | 116 | if (!(remv[nblock] & (1ULL << inblock))) { 117 | keep_out[num_to_keep++] = i; 118 | unsigned long long *p = &mask_host[0] + i * col_blocks; 119 | for (int j = nblock; j < col_blocks; j++) { 120 | remv[j] |= p[j]; 121 | } 122 | } 123 | } 124 | 125 | THCudaFree(state, mask_dev); 126 | // TODO improve this part 127 | return std::get<0>(order_t.index({ 128 | keep.narrow(/*dim=*/0, /*start=*/0, /*length=*/num_to_keep).to( 129 | order_t.device(), keep.scalar_type()) 130 | }).sort(0, false)); 131 | } -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/nms/ext/cuda/vision.h: -------------------------------------------------------------------------------- 1 | // Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | #pragma once 3 | #include 4 | 5 | at::Tensor nms_cuda(const at::Tensor boxes, float nms_overlap_thresh); 6 | 7 | 8 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/nms/ext/nms.h: -------------------------------------------------------------------------------- 1 | // Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | #pragma once 3 | #include "cpu/vision.h" 4 | 5 | #ifdef WITH_CUDA 6 | #include "cuda/vision.h" 7 | #endif 8 | 9 | 10 | at::Tensor nms(const at::Tensor& dets, 11 | const at::Tensor& scores, 12 | const float threshold) { 13 | 14 | if (dets.type().is_cuda()) { 15 | #ifdef WITH_CUDA 16 | // TODO raise error if not compiled with CUDA 17 | if (dets.numel() == 0) 18 | return at::empty({0}, dets.options().dtype(at::kLong).device(at::kCPU)); 19 | auto b = at::cat({dets, scores.unsqueeze(1)}, 1); 20 | return nms_cuda(b, threshold); 21 | #else 22 | AT_ERROR("Not compiled with GPU support"); 23 | #endif 24 | } 25 | 26 | at::Tensor result = nms_cpu(dets, scores, threshold); 27 | return result; 28 | } 29 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/nms/ext/vision.cpp: -------------------------------------------------------------------------------- 1 | // Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | #include "nms.h" 3 | 4 | 5 | PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) { 6 | m.def("nms", &nms, "non-maximum suppression"); 7 | } 8 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/nms/nms.py: -------------------------------------------------------------------------------- 1 | import warnings 2 | import torchvision 3 | 4 | try: 5 | import torch 6 | import torch_extension 7 | 8 | _nms = torch_extension.nms 9 | except ImportError: 10 | if torchvision.__version__ >= '0.3.0': 11 | _nms = torchvision.ops.nms 12 | else: 13 | from .python_nms import python_nms 14 | 15 | _nms = python_nms 16 | warnings.warn('You are using python version NMS, which is very very slow. Try compile c++ NMS ' 17 | 'using `cd ext & python build.py build_ext develop`') 18 | 19 | 20 | def boxes_nms(boxes, scores, nms_thresh, max_count=-1): 21 | """ Performs non-maximum suppression, run on GPU or CPU according to 22 | boxes's device. 23 | Args: 24 | boxes(Tensor): `xyxy` mode boxes, use absolute coordinates(or relative coordinates), shape is (n, 4) 25 | scores(Tensor): scores, shape is (n, ) 26 | nms_thresh(float): thresh 27 | max_count (int): if > 0, then only the top max_proposals are kept after non-maximum suppression 28 | Returns: 29 | indices kept. 30 | """ 31 | keep = _nms(boxes, scores, nms_thresh) 32 | if max_count > 0: 33 | keep = keep[:max_count] 34 | return keep 35 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/nms/python_nms.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import numpy as np 3 | 4 | 5 | def python_nms(boxes, scores, nms_thresh): 6 | """ Performs non-maximum suppression using numpy 7 | Args: 8 | boxes(Tensor): `xyxy` mode boxes, use absolute coordinates(not support relative coordinates), 9 | shape is (n, 4) 10 | scores(Tensor): scores, shape is (n, ) 11 | nms_thresh(float): thresh 12 | Returns: 13 | indices kept. 14 | """ 15 | if boxes.numel() == 0: 16 | return torch.empty((0,), dtype=torch.long) 17 | # Use numpy to run nms. Running nms in PyTorch code on CPU is really slow. 18 | origin_device = boxes.device 19 | cpu_device = torch.device('cpu') 20 | boxes = boxes.to(cpu_device).numpy() 21 | scores = scores.to(cpu_device).numpy() 22 | 23 | x1 = boxes[:, 0] 24 | y1 = boxes[:, 1] 25 | x2 = boxes[:, 2] 26 | y2 = boxes[:, 3] 27 | areas = (x2 - x1) * (y2 - y1) 28 | order = np.argsort(scores)[::-1] 29 | num_detections = boxes.shape[0] 30 | suppressed = np.zeros((num_detections,), dtype=np.bool) 31 | for _i in range(num_detections): 32 | i = order[_i] 33 | if suppressed[i]: 34 | continue 35 | ix1 = x1[i] 36 | iy1 = y1[i] 37 | ix2 = x2[i] 38 | iy2 = y2[i] 39 | iarea = areas[i] 40 | 41 | for _j in range(_i + 1, num_detections): 42 | j = order[_j] 43 | if suppressed[j]: 44 | continue 45 | 46 | xx1 = max(ix1, x1[j]) 47 | yy1 = max(iy1, y1[j]) 48 | xx2 = min(ix2, x2[j]) 49 | yy2 = min(iy2, y2[j]) 50 | w = max(0, xx2 - xx1) 51 | h = max(0, yy2 - yy1) 52 | 53 | inter = w * h 54 | ovr = inter / (iarea + areas[j] - inter) 55 | if ovr >= nms_thresh: 56 | suppressed[j] = True 57 | keep = np.nonzero(suppressed == 0)[0] 58 | keep = torch.from_numpy(keep).to(origin_device) 59 | return keep 60 | -------------------------------------------------------------------------------- /deep_sort/detector/YOLOv3/weight/.gitkeep: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/deep_sort/detector/YOLOv3/weight/.gitkeep -------------------------------------------------------------------------------- /deep_sort/detector/__init__.py: -------------------------------------------------------------------------------- 1 | from .YOLOv3 import YOLOv3 2 | 3 | 4 | __all__ = ['build_detector'] 5 | 6 | def build_detector(cfg, use_cuda): 7 | return YOLOv3(cfg.YOLOV3.CFG, cfg.YOLOV3.WEIGHT, cfg.YOLOV3.CLASS_NAMES, 8 | score_thresh=cfg.YOLOV3.SCORE_THRESH, nms_thresh=cfg.YOLOV3.NMS_THRESH, 9 | is_xywh=True, use_cuda=use_cuda) 10 | -------------------------------------------------------------------------------- /deep_sort/ped_det_server.py: -------------------------------------------------------------------------------- 1 | """ 2 | This module gets video in input and outputs the 3 | json file with coordination of bboxes in the video. 4 | 5 | """ 6 | from os.path import basename, splitext, join, isfile, isdir, dirname 7 | from os import makedirs 8 | 9 | from tqdm import tqdm 10 | import cv2 11 | import argparse 12 | import torch 13 | 14 | from detector import build_detector 15 | from deep_sort import build_tracker 16 | from utils.tools import tik_tok, is_video 17 | from utils.draw import compute_color_for_labels 18 | from utils.parser import get_config 19 | from utils.json_logger import BboxToJsonLogger 20 | import warnings 21 | 22 | 23 | def parse_args(): 24 | parser = argparse.ArgumentParser() 25 | parser.add_argument("--VIDEO_PATH", type=str, default="./demo/ped.avi") 26 | parser.add_argument("--config_detection", type=str, default="./configs/yolov3.yaml") 27 | parser.add_argument("--config_deepsort", type=str, default="./configs/deep_sort.yaml") 28 | parser.add_argument("--write-fps", type=int, default=20) 29 | parser.add_argument("--frame_interval", type=int, default=1) 30 | parser.add_argument("--save_path", type=str, default="./output") 31 | parser.add_argument("--cpu", dest="use_cuda", action="store_false", default=True) 32 | args = parser.parse_args() 33 | 34 | assert isfile(args.VIDEO_PATH), "Error: Video not found" 35 | assert is_video(args.VIDEO_PATH), "Error: Not Supported format" 36 | if args.frame_interval < 1: args.frame_interval = 1 37 | 38 | return args 39 | 40 | 41 | class VideoTracker(object): 42 | def __init__(self, cfg, args): 43 | self.cfg = cfg 44 | self.args = args 45 | use_cuda = args.use_cuda and torch.cuda.is_available() 46 | if not use_cuda: 47 | warnings.warn("Running in cpu mode!") 48 | 49 | self.vdo = cv2.VideoCapture() 50 | self.detector = build_detector(cfg, use_cuda=use_cuda) 51 | self.deepsort = build_tracker(cfg, use_cuda=use_cuda) 52 | self.class_names = self.detector.class_names 53 | 54 | # Configure output video and json 55 | self.logger = BboxToJsonLogger() 56 | filename, extension = splitext(basename(self.args.VIDEO_PATH)) 57 | self.output_file = join(self.args.save_path, f'{filename}.avi') 58 | self.json_output = join(self.args.save_path, f'{filename}.json') 59 | if not isdir(dirname(self.json_output)): 60 | makedirs(dirname(self.json_output)) 61 | 62 | def __enter__(self): 63 | self.vdo.open(self.args.VIDEO_PATH) 64 | self.total_frames = int(cv2.VideoCapture.get(self.vdo, cv2.CAP_PROP_FRAME_COUNT)) 65 | self.im_width = int(self.vdo.get(cv2.CAP_PROP_FRAME_WIDTH)) 66 | self.im_height = int(self.vdo.get(cv2.CAP_PROP_FRAME_HEIGHT)) 67 | 68 | video_details = {'frame_width': self.im_width, 69 | 'frame_height': self.im_height, 70 | 'frame_rate': self.args.write_fps, 71 | 'video_name': self.args.VIDEO_PATH} 72 | codec = cv2.VideoWriter_fourcc(*'XVID') 73 | self.writer = cv2.VideoWriter(self.output_file, codec, self.args.write_fps, 74 | (self.im_width, self.im_height)) 75 | self.logger.add_video_details(**video_details) 76 | 77 | assert self.vdo.isOpened() 78 | return self 79 | 80 | def __exit__(self, exc_type, exc_value, exc_traceback): 81 | if exc_type: 82 | print(exc_type, exc_value, exc_traceback) 83 | 84 | def run(self): 85 | idx_frame = 0 86 | pbar = tqdm(total=self.total_frames + 1) 87 | while self.vdo.grab(): 88 | if idx_frame % args.frame_interval == 0: 89 | _, ori_im = self.vdo.retrieve() 90 | timestamp = self.vdo.get(cv2.CAP_PROP_POS_MSEC) 91 | frame_id = int(self.vdo.get(cv2.CAP_PROP_POS_FRAMES)) 92 | self.logger.add_frame(frame_id=frame_id, timestamp=timestamp) 93 | self.detection(frame=ori_im, frame_id=frame_id) 94 | self.save_frame(ori_im) 95 | idx_frame += 1 96 | pbar.update() 97 | self.logger.json_output(self.json_output) 98 | 99 | @tik_tok 100 | def detection(self, frame, frame_id): 101 | im = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) 102 | # do detection 103 | bbox_xywh, cls_conf, cls_ids = self.detector(im) 104 | if bbox_xywh is not None: 105 | # select person class 106 | mask = cls_ids == 0 107 | 108 | bbox_xywh = bbox_xywh[mask] 109 | bbox_xywh[:, 3:] *= 1.2 # bbox dilation just in case bbox too small 110 | cls_conf = cls_conf[mask] 111 | 112 | # do tracking 113 | outputs = self.deepsort.update(bbox_xywh, cls_conf, im) 114 | 115 | # draw boxes for visualization 116 | if len(outputs) > 0: 117 | frame = self.draw_boxes(img=frame, frame_id=frame_id, output=outputs) 118 | 119 | def draw_boxes(self, img, frame_id, output, offset=(0, 0)): 120 | for i, box in enumerate(output): 121 | x1, y1, x2, y2, identity = [int(ii) for ii in box] 122 | self.logger.add_bbox_to_frame(frame_id=frame_id, 123 | bbox_id=identity, 124 | top=y1, 125 | left=x1, 126 | width=x2 - x1, 127 | height=y2 - y1) 128 | x1 += offset[0] 129 | x2 += offset[0] 130 | y1 += offset[1] 131 | y2 += offset[1] 132 | 133 | # box text and bar 134 | self.logger.add_label_to_bbox(frame_id=frame_id, bbox_id=identity, category='pedestrian', confidence=0.9) 135 | color = compute_color_for_labels(identity) 136 | label = '{}{:d}'.format("", identity) 137 | t_size = cv2.getTextSize(label, cv2.FONT_HERSHEY_PLAIN, 2, 2)[0] 138 | cv2.rectangle(img, (x1, y1), (x2, y2), color, 3) 139 | cv2.rectangle(img, (x1, y1), (x1 + t_size[0] + 3, y1 + t_size[1] + 4), color, -1) 140 | cv2.putText(img, label, (x1, y1 + t_size[1] + 4), cv2.FONT_HERSHEY_PLAIN, 2, [255, 255, 255], 2) 141 | return img 142 | 143 | def save_frame(self, frame) -> None: 144 | if frame is not None: self.writer.write(frame) 145 | 146 | 147 | if __name__ == "__main__": 148 | args = parse_args() 149 | cfg = get_config() 150 | cfg.merge_from_file(args.config_detection) 151 | cfg.merge_from_file(args.config_deepsort) 152 | 153 | with VideoTracker(cfg, args) as vdo_trk: 154 | vdo_trk.run() 155 | 156 | -------------------------------------------------------------------------------- /deep_sort/scripts/yolov3_deepsort.sh: -------------------------------------------------------------------------------- 1 | python yolov3_deepsort.py [VIDEO_PATH] --config_detection -------------------------------------------------------------------------------- /deep_sort/scripts/yolov3_tiny_deepsort.sh: -------------------------------------------------------------------------------- 1 | python yolov3_deepsort.py [VIDEO_PATH] --config_detection ./configs/yolov3_tiny.yaml -------------------------------------------------------------------------------- /deep_sort/utils/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/deep_sort/utils/__init__.py -------------------------------------------------------------------------------- /deep_sort/utils/asserts.py: -------------------------------------------------------------------------------- 1 | from os import environ 2 | 3 | 4 | def assert_in(file, files_to_check): 5 | if file not in files_to_check: 6 | raise AssertionError("{} does not exist in the list".format(str(file))) 7 | return True 8 | 9 | 10 | def assert_in_env(check_list: list): 11 | for item in check_list: 12 | assert_in(item, environ.keys()) 13 | return True 14 | -------------------------------------------------------------------------------- /deep_sort/utils/draw.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import cv2 3 | 4 | palette = (2 ** 11 - 1, 2 ** 15 - 1, 2 ** 20 - 1) 5 | 6 | 7 | def compute_color_for_labels(label): 8 | """ 9 | Simple function that adds fixed color depending on the class 10 | """ 11 | color = [int((p * (label ** 2 - label + 1)) % 255) for p in palette] 12 | return tuple(color) 13 | 14 | 15 | def draw_boxes(img, bbox, identities=None, offset=(0,0)): 16 | for i,box in enumerate(bbox): 17 | x1,y1,x2,y2 = [int(i) for i in box] 18 | x1 += offset[0] 19 | x2 += offset[0] 20 | y1 += offset[1] 21 | y2 += offset[1] 22 | # box text and bar 23 | id = int(identities[i]) if identities is not None else 0 24 | color = compute_color_for_labels(id) 25 | label = '{}{:d}'.format("", id) 26 | t_size = cv2.getTextSize(label, cv2.FONT_HERSHEY_PLAIN, 2 , 2)[0] 27 | cv2.rectangle(img,(x1, y1),(x2,y2),color,3) 28 | cv2.rectangle(img,(x1, y1),(x1+t_size[0]+3,y1+t_size[1]+4), color,-1) 29 | cv2.putText(img,label,(x1,y1+t_size[1]+4), cv2.FONT_HERSHEY_PLAIN, 2, [255,255,255], 2) 30 | return img 31 | 32 | 33 | 34 | if __name__ == '__main__': 35 | for i in range(82): 36 | print(compute_color_for_labels(i)) 37 | -------------------------------------------------------------------------------- /deep_sort/utils/evaluation.py: -------------------------------------------------------------------------------- 1 | import os 2 | import numpy as np 3 | import copy 4 | import motmetrics as mm 5 | mm.lap.default_solver = 'lap' 6 | from utils.io import read_results, unzip_objs 7 | 8 | 9 | class Evaluator(object): 10 | 11 | def __init__(self, data_root, seq_name, data_type): 12 | self.data_root = data_root 13 | self.seq_name = seq_name 14 | self.data_type = data_type 15 | 16 | self.load_annotations() 17 | self.reset_accumulator() 18 | 19 | def load_annotations(self): 20 | assert self.data_type == 'mot' 21 | 22 | gt_filename = os.path.join(self.data_root, self.seq_name, 'gt', 'gt.txt') 23 | self.gt_frame_dict = read_results(gt_filename, self.data_type, is_gt=True) 24 | self.gt_ignore_frame_dict = read_results(gt_filename, self.data_type, is_ignore=True) 25 | 26 | def reset_accumulator(self): 27 | self.acc = mm.MOTAccumulator(auto_id=True) 28 | 29 | def eval_frame(self, frame_id, trk_tlwhs, trk_ids, rtn_events=False): 30 | # results 31 | trk_tlwhs = np.copy(trk_tlwhs) 32 | trk_ids = np.copy(trk_ids) 33 | 34 | # gts 35 | gt_objs = self.gt_frame_dict.get(frame_id, []) 36 | gt_tlwhs, gt_ids = unzip_objs(gt_objs)[:2] 37 | 38 | # ignore boxes 39 | ignore_objs = self.gt_ignore_frame_dict.get(frame_id, []) 40 | ignore_tlwhs = unzip_objs(ignore_objs)[0] 41 | 42 | 43 | # remove ignored results 44 | keep = np.ones(len(trk_tlwhs), dtype=bool) 45 | iou_distance = mm.distances.iou_matrix(ignore_tlwhs, trk_tlwhs, max_iou=0.5) 46 | if len(iou_distance) > 0: 47 | match_is, match_js = mm.lap.linear_sum_assignment(iou_distance) 48 | match_is, match_js = map(lambda a: np.asarray(a, dtype=int), [match_is, match_js]) 49 | match_ious = iou_distance[match_is, match_js] 50 | 51 | match_js = np.asarray(match_js, dtype=int) 52 | match_js = match_js[np.logical_not(np.isnan(match_ious))] 53 | keep[match_js] = False 54 | trk_tlwhs = trk_tlwhs[keep] 55 | trk_ids = trk_ids[keep] 56 | 57 | # get distance matrix 58 | iou_distance = mm.distances.iou_matrix(gt_tlwhs, trk_tlwhs, max_iou=0.5) 59 | 60 | # acc 61 | self.acc.update(gt_ids, trk_ids, iou_distance) 62 | 63 | if rtn_events and iou_distance.size > 0 and hasattr(self.acc, 'last_mot_events'): 64 | events = self.acc.last_mot_events # only supported by https://github.com/longcw/py-motmetrics 65 | else: 66 | events = None 67 | return events 68 | 69 | def eval_file(self, filename): 70 | self.reset_accumulator() 71 | 72 | result_frame_dict = read_results(filename, self.data_type, is_gt=False) 73 | frames = sorted(list(set(self.gt_frame_dict.keys()) | set(result_frame_dict.keys()))) 74 | for frame_id in frames: 75 | trk_objs = result_frame_dict.get(frame_id, []) 76 | trk_tlwhs, trk_ids = unzip_objs(trk_objs)[:2] 77 | self.eval_frame(frame_id, trk_tlwhs, trk_ids, rtn_events=False) 78 | 79 | return self.acc 80 | 81 | @staticmethod 82 | def get_summary(accs, names, metrics=('mota', 'num_switches', 'idp', 'idr', 'idf1', 'precision', 'recall')): 83 | names = copy.deepcopy(names) 84 | if metrics is None: 85 | metrics = mm.metrics.motchallenge_metrics 86 | metrics = copy.deepcopy(metrics) 87 | 88 | mh = mm.metrics.create() 89 | summary = mh.compute_many( 90 | accs, 91 | metrics=metrics, 92 | names=names, 93 | generate_overall=True 94 | ) 95 | 96 | return summary 97 | 98 | @staticmethod 99 | def save_summary(summary, filename): 100 | import pandas as pd 101 | writer = pd.ExcelWriter(filename) 102 | summary.to_excel(writer) 103 | writer.save() 104 | -------------------------------------------------------------------------------- /deep_sort/utils/io.py: -------------------------------------------------------------------------------- 1 | import os 2 | from typing import Dict 3 | import numpy as np 4 | 5 | # from utils.log import get_logger 6 | 7 | 8 | def write_results(filename, results, data_type): 9 | if data_type == 'mot': 10 | save_format = '{frame},{id},{x1},{y1},{w},{h},-1,-1,-1,-1\n' 11 | elif data_type == 'kitti': 12 | save_format = '{frame} {id} pedestrian 0 0 -10 {x1} {y1} {x2} {y2} -10 -10 -10 -1000 -1000 -1000 -10\n' 13 | else: 14 | raise ValueError(data_type) 15 | 16 | with open(filename, 'w') as f: 17 | for frame_id, tlwhs, track_ids in results: 18 | if data_type == 'kitti': 19 | frame_id -= 1 20 | for tlwh, track_id in zip(tlwhs, track_ids): 21 | if track_id < 0: 22 | continue 23 | x1, y1, w, h = tlwh 24 | x2, y2 = x1 + w, y1 + h 25 | line = save_format.format(frame=frame_id, id=track_id, x1=x1, y1=y1, x2=x2, y2=y2, w=w, h=h) 26 | f.write(line) 27 | 28 | 29 | # def write_results(filename, results_dict: Dict, data_type: str): 30 | # if not filename: 31 | # return 32 | # path = os.path.dirname(filename) 33 | # if not os.path.exists(path): 34 | # os.makedirs(path) 35 | 36 | # if data_type in ('mot', 'mcmot', 'lab'): 37 | # save_format = '{frame},{id},{x1},{y1},{w},{h},1,-1,-1,-1\n' 38 | # elif data_type == 'kitti': 39 | # save_format = '{frame} {id} pedestrian -1 -1 -10 {x1} {y1} {x2} {y2} -1 -1 -1 -1000 -1000 -1000 -10 {score}\n' 40 | # else: 41 | # raise ValueError(data_type) 42 | 43 | # with open(filename, 'w') as f: 44 | # for frame_id, frame_data in results_dict.items(): 45 | # if data_type == 'kitti': 46 | # frame_id -= 1 47 | # for tlwh, track_id in frame_data: 48 | # if track_id < 0: 49 | # continue 50 | # x1, y1, w, h = tlwh 51 | # x2, y2 = x1 + w, y1 + h 52 | # line = save_format.format(frame=frame_id, id=track_id, x1=x1, y1=y1, x2=x2, y2=y2, w=w, h=h, score=1.0) 53 | # f.write(line) 54 | # logger.info('Save results to {}'.format(filename)) 55 | 56 | 57 | def read_results(filename, data_type: str, is_gt=False, is_ignore=False): 58 | if data_type in ('mot', 'lab'): 59 | read_fun = read_mot_results 60 | else: 61 | raise ValueError('Unknown data type: {}'.format(data_type)) 62 | 63 | return read_fun(filename, is_gt, is_ignore) 64 | 65 | 66 | """ 67 | labels={'ped', ... % 1 68 | 'person_on_vhcl', ... % 2 69 | 'car', ... % 3 70 | 'bicycle', ... % 4 71 | 'mbike', ... % 5 72 | 'non_mot_vhcl', ... % 6 73 | 'static_person', ... % 7 74 | 'distractor', ... % 8 75 | 'occluder', ... % 9 76 | 'occluder_on_grnd', ... %10 77 | 'occluder_full', ... % 11 78 | 'reflection', ... % 12 79 | 'crowd' ... % 13 80 | }; 81 | """ 82 | 83 | 84 | def read_mot_results(filename, is_gt, is_ignore): 85 | valid_labels = {1} 86 | ignore_labels = {2, 7, 8, 12} 87 | results_dict = dict() 88 | if os.path.isfile(filename): 89 | with open(filename, 'r') as f: 90 | for line in f.readlines(): 91 | linelist = line.split(',') 92 | if len(linelist) < 7: 93 | continue 94 | fid = int(linelist[0]) 95 | if fid < 1: 96 | continue 97 | results_dict.setdefault(fid, list()) 98 | 99 | if is_gt: 100 | if 'MOT16-' in filename or 'MOT17-' in filename: 101 | label = int(float(linelist[7])) 102 | mark = int(float(linelist[6])) 103 | if mark == 0 or label not in valid_labels: 104 | continue 105 | score = 1 106 | elif is_ignore: 107 | if 'MOT16-' in filename or 'MOT17-' in filename: 108 | label = int(float(linelist[7])) 109 | vis_ratio = float(linelist[8]) 110 | if label not in ignore_labels and vis_ratio >= 0: 111 | continue 112 | else: 113 | continue 114 | score = 1 115 | else: 116 | score = float(linelist[6]) 117 | 118 | tlwh = tuple(map(float, linelist[2:6])) 119 | target_id = int(linelist[1]) 120 | 121 | results_dict[fid].append((tlwh, target_id, score)) 122 | 123 | return results_dict 124 | 125 | 126 | def unzip_objs(objs): 127 | if len(objs) > 0: 128 | tlwhs, ids, scores = zip(*objs) 129 | else: 130 | tlwhs, ids, scores = [], [], [] 131 | tlwhs = np.asarray(tlwhs, dtype=float).reshape(-1, 4) 132 | 133 | return tlwhs, ids, scores -------------------------------------------------------------------------------- /deep_sort/utils/log.py: -------------------------------------------------------------------------------- 1 | import logging 2 | 3 | 4 | def get_logger(name='root'): 5 | formatter = logging.Formatter( 6 | # fmt='%(asctime)s [%(levelname)s]: %(filename)s(%(funcName)s:%(lineno)s) >> %(message)s') 7 | fmt='%(asctime)s [%(levelname)s]: %(message)s', datefmt='%Y-%m-%d %H:%M:%S') 8 | 9 | handler = logging.StreamHandler() 10 | handler.setFormatter(formatter) 11 | 12 | logger = logging.getLogger(name) 13 | logger.setLevel(logging.INFO) 14 | logger.addHandler(handler) 15 | return logger 16 | 17 | 18 | -------------------------------------------------------------------------------- /deep_sort/utils/parser.py: -------------------------------------------------------------------------------- 1 | import os 2 | import yaml 3 | from easydict import EasyDict as edict 4 | 5 | class YamlParser(edict): 6 | """ 7 | This is yaml parser based on EasyDict. 8 | """ 9 | def __init__(self, cfg_dict=None, config_file=None): 10 | if cfg_dict is None: 11 | cfg_dict = {} 12 | 13 | if config_file is not None: 14 | assert(os.path.isfile(config_file)) 15 | with open(config_file, 'r') as fo: 16 | cfg_dict.update(yaml.load(fo.read())) 17 | 18 | super(YamlParser, self).__init__(cfg_dict) 19 | 20 | 21 | def merge_from_file(self, config_file): 22 | with open(config_file, 'r') as fo: 23 | self.update(yaml.load(fo.read())) 24 | 25 | 26 | def merge_from_dict(self, config_dict): 27 | self.update(config_dict) 28 | 29 | 30 | def get_config(config_file=None): 31 | return YamlParser(config_file=config_file) 32 | 33 | 34 | if __name__ == "__main__": 35 | cfg = YamlParser(config_file="../configs/yolov3.yaml") 36 | cfg.merge_from_file("../configs/deep_sort.yaml") 37 | 38 | import ipdb; ipdb.set_trace() -------------------------------------------------------------------------------- /deep_sort/utils/tools.py: -------------------------------------------------------------------------------- 1 | from functools import wraps 2 | from time import time 3 | 4 | 5 | def is_video(ext: str): 6 | """ 7 | Returns true if ext exists in 8 | allowed_exts for video files. 9 | 10 | Args: 11 | ext: 12 | 13 | Returns: 14 | 15 | """ 16 | 17 | allowed_exts = ('.mp4', '.webm', '.ogg', '.avi', '.wmv', '.mkv', '.3gp') 18 | return any((ext.endswith(x) for x in allowed_exts)) 19 | 20 | 21 | def tik_tok(func): 22 | """ 23 | keep track of time for each process. 24 | Args: 25 | func: 26 | 27 | Returns: 28 | 29 | """ 30 | @wraps(func) 31 | def _time_it(*args, **kwargs): 32 | start = time() 33 | try: 34 | return func(*args, **kwargs) 35 | finally: 36 | end_ = time() 37 | print("time: {:.03f}s, fps: {:.03f}".format(end_ - start, 1 / (end_ - start))) 38 | 39 | return _time_it 40 | -------------------------------------------------------------------------------- /deep_sort/webserver/.env: -------------------------------------------------------------------------------- 1 | project_root="C:\Users\ZQ_deep_sort_pytorch" 2 | model_type="yolov3" 3 | output_dir="public/" 4 | json_output="json_output/" # ignored for the moment in ped_det_online_server.py 5 | reid_ckpt="deep_sort/deep/checkpoint/ckpt.t7" 6 | yolov3_cfg="detector/YOLOv3/cfg/yolo_v3.cfg" 7 | yolov3_weight="detector/YOLOv3/weight/yolov3.weights" 8 | yolov3_tiny_cfg="detector/YOLOv3/cfg/yolov3-tiny.cfg" 9 | yolov3_tiny_weight="detector/YOLOv3/weight/yolov3-tiny.weights" 10 | yolov3_class_names="detector/YOLOv3/cfg/coco.names" 11 | analysis_output="video_analysis/" 12 | app="flask_stream_server.py" 13 | camera_stream= "rtsp://user@111.222.333.444:somesecretcode" 14 | -------------------------------------------------------------------------------- /deep_sort/webserver/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/deep_sort/webserver/__init__.py -------------------------------------------------------------------------------- /deep_sort/webserver/config/config.py: -------------------------------------------------------------------------------- 1 | import os 2 | 3 | app_dir = os.path.abspath(os.path.dirname(__file__)) 4 | 5 | 6 | class BaseConfig: 7 | SECRET_KEY = os.environ.get('SECRET_KEY') or 'Sm9obiBTY2hyb20ga2lja3MgYXNz' 8 | SERVER_NAME = '127.0.0.1:8888' 9 | 10 | 11 | class DevelopmentConfig(BaseConfig): 12 | ENV = 'development' 13 | DEBUG = True 14 | 15 | 16 | class TestingConfig(BaseConfig): 17 | DEBUG = True 18 | 19 | 20 | class ProductionConfig(BaseConfig): 21 | DEBUG = False 22 | -------------------------------------------------------------------------------- /deep_sort/webserver/images/Thumbs.db: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/deep_sort/webserver/images/Thumbs.db -------------------------------------------------------------------------------- /deep_sort/webserver/images/arc.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/deep_sort/webserver/images/arc.png -------------------------------------------------------------------------------- /deep_sort/webserver/images/request.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/deep_sort/webserver/images/request.png -------------------------------------------------------------------------------- /deep_sort/webserver/readme.md: -------------------------------------------------------------------------------- 1 | # Stream pedestrian detection web server 2 | 3 | ### Requirements 4 | 5 | - python = 3.7 6 | - redis 7 | - flask 8 | - opencv 9 | - pytorch 10 | - dotenv 11 | 12 | Please note that you need to install redis on your system. 13 | 14 | ### The architecture. 15 | 16 | ![web server architecture](images/arc.png) 17 | 18 | 1 - `RealTimeTracking` reads frames from rtsp link using threads 19 | (Using threads make the web server robust against network packet loss) 20 | 21 | 2 - In `RealTimeTracking.run` function in each iteration frame is stored in redis cache on server. 22 | 23 | 3 - Now we can serve the frames on redis to clients. 24 | 25 | 4 - To start the pedestrian detection, after running 26 | `rtsp_webserver.py`, send a GET request on `127.0.0.1:8888/run` 27 | with setting these GET method parameters 28 | 29 | | Param | Value | Description | 30 | | :-------------: | :-------------: | :-------------: | 31 | | run | 1/0 | to start the tracking set 1/ to stop tracking service set it as 0| 32 | | camera_stream | 'rtsp://ip:port/admin...' | provide it with valid rtsp link | 33 | 34 | for example: 35 | 36 | (to start the service) 127.0.0.1:8888/run?run=1 37 | (to stop the service) 127.0.0.1:8888/run?run=0 38 | (to change the camera) 39 | 1- 127.0.0.1:8888/run?run=0 (first stop the current service) 40 | 2- 127.0.0.1:8888/run?run=0&camera_stream=rtsp://ip:port/admin... (then start it with another rtsp link) 41 | 42 | ![web server architecture](images/request.png) 43 | 44 | 45 | 5 - get pedestrian detection stream in `127.0.0.1:8888` 46 | -------------------------------------------------------------------------------- /deep_sort/webserver/rtsp_threaded_tracker.py: -------------------------------------------------------------------------------- 1 | import warnings 2 | from os import getenv 3 | import sys 4 | from os.path import dirname, abspath 5 | 6 | sys.path.append(dirname(dirname(abspath(__file__)))) 7 | 8 | import torch 9 | from deep_sort import build_tracker 10 | from detector import build_detector 11 | import cv2 12 | from utils.draw import compute_color_for_labels 13 | from concurrent.futures import ThreadPoolExecutor 14 | from redis import Redis 15 | 16 | redis_cache = Redis('127.0.0.1') 17 | 18 | 19 | class RealTimeTracking(object): 20 | """ 21 | This class is built to get frame from rtsp link and continuously 22 | assign each frame to an attribute namely as frame in order to 23 | compensate the network packet loss. then we use flask to give it 24 | as service to client. 25 | Args: 26 | args: parse_args inputs 27 | cfg: deepsort dict and yolo-model cfg from server_cfg file 28 | 29 | """ 30 | 31 | def __init__(self, cfg, args): 32 | # Create a VideoCapture object 33 | self.cfg = cfg 34 | self.args = args 35 | use_cuda = self.args.use_cuda and torch.cuda.is_available() 36 | 37 | if not use_cuda: 38 | warnings.warn(UserWarning("Running in cpu mode!")) 39 | 40 | self.detector = build_detector(cfg, use_cuda=use_cuda) 41 | self.deepsort = build_tracker(cfg, use_cuda=use_cuda) 42 | self.class_names = self.detector.class_names 43 | 44 | self.vdo = cv2.VideoCapture(self.args.input) 45 | self.status, self.frame = None, None 46 | self.total_frames = int(cv2.VideoCapture.get(self.vdo, cv2.CAP_PROP_FRAME_COUNT)) 47 | self.im_width = int(self.vdo.get(cv2.CAP_PROP_FRAME_WIDTH)) 48 | self.im_height = int(self.vdo.get(cv2.CAP_PROP_FRAME_HEIGHT)) 49 | 50 | self.output_frame = None 51 | 52 | self.thread = ThreadPoolExecutor(max_workers=1) 53 | self.thread.submit(self.update) 54 | 55 | def update(self): 56 | while True: 57 | if self.vdo.isOpened(): 58 | (self.status, self.frame) = self.vdo.read() 59 | 60 | def run(self): 61 | print('streaming started ...') 62 | while getenv('in_progress') != 'off': 63 | try: 64 | frame = self.frame.copy() 65 | self.detection(frame=frame) 66 | frame_to_bytes = cv2.imencode('.jpg', frame)[1].tobytes() 67 | redis_cache.set('frame', frame_to_bytes) 68 | except AttributeError: 69 | pass 70 | print('streaming stopped ...') 71 | 72 | 73 | def detection(self, frame): 74 | im = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) 75 | # do detection 76 | bbox_xywh, cls_conf, cls_ids = self.detector(im) 77 | if bbox_xywh is not None: 78 | # select person class 79 | mask = cls_ids == 0 80 | 81 | bbox_xywh = bbox_xywh[mask] 82 | bbox_xywh[:, 3:] *= 1.2 # bbox dilation just in case bbox too small 83 | cls_conf = cls_conf[mask] 84 | 85 | # do tracking 86 | outputs = self.deepsort.update(bbox_xywh, cls_conf, im) 87 | 88 | # draw boxes for visualization 89 | if len(outputs) > 0: 90 | self.draw_boxes(img=frame, output=outputs) 91 | 92 | @staticmethod 93 | def draw_boxes(img, output, offset=(0, 0)): 94 | for i, box in enumerate(output): 95 | x1, y1, x2, y2, identity = [int(ii) for ii in box] 96 | x1 += offset[0] 97 | x2 += offset[0] 98 | y1 += offset[1] 99 | y2 += offset[1] 100 | 101 | # box text and bar 102 | color = compute_color_for_labels(identity) 103 | label = '{}{:d}'.format("", identity) 104 | t_size = cv2.getTextSize(label, cv2.FONT_HERSHEY_PLAIN, 2, 2)[0] 105 | cv2.rectangle(img, (x1, y1), (x2, y2), color, 3) 106 | cv2.rectangle(img, (x1, y1), (x1 + t_size[0] + 3, y1 + t_size[1] + 4), color, -1) 107 | cv2.putText(img, label, (x1, y1 + t_size[1] + 4), cv2.FONT_HERSHEY_PLAIN, 2, [255, 255, 255], 2) 108 | return img 109 | -------------------------------------------------------------------------------- /deep_sort/webserver/rtsp_webserver.py: -------------------------------------------------------------------------------- 1 | """ 2 | 3 | # TODO: Load ML model with redis and keep it for sometime. 4 | 1- detector/yolov3/detector.py |=> yolov3 weightfile -> redis cache 5 | 2- deepsort/deep/feature_extractor |=> model_path -> redis cache 6 | 3- Use tmpfs (Insert RAM as a virtual disk and store model state): https://pypi.org/project/memory-tempfile/ 7 | 8 | """ 9 | from os.path import join 10 | from os import getenv, environ 11 | from dotenv import load_dotenv 12 | import argparse 13 | from threading import Thread 14 | 15 | from redis import Redis 16 | from flask import Response, Flask, jsonify, request, abort 17 | 18 | from rtsp_threaded_tracker import RealTimeTracking 19 | from server_cfg import model, deep_sort_dict 20 | from config.config import DevelopmentConfig 21 | from utils.parser import get_config 22 | 23 | redis_cache = Redis('127.0.0.1') 24 | app = Flask(__name__) 25 | environ['in_progress'] = 'off' 26 | 27 | 28 | def parse_args(): 29 | """ 30 | Parses the arguments 31 | Returns: 32 | argparse Namespace 33 | """ 34 | assert 'project_root' in environ.keys() 35 | project_root = getenv('project_root') 36 | parser = argparse.ArgumentParser() 37 | 38 | parser.add_argument("--input", 39 | type=str, 40 | default=getenv('camera_stream')) 41 | 42 | parser.add_argument("--model", 43 | type=str, 44 | default=join(project_root, 45 | getenv('model_type'))) 46 | 47 | parser.add_argument("--cpu", 48 | dest="use_cuda", 49 | action="store_false", default=True) 50 | args = parser.parse_args() 51 | 52 | return args 53 | 54 | 55 | def gen(): 56 | """ 57 | 58 | Returns: video frames from redis cache 59 | 60 | """ 61 | while True: 62 | frame = redis_cache.get('frame') 63 | if frame is not None: 64 | yield b'--frame\r\n'b'Content-Type: image/jpeg\r\n\r\n' + frame + b'\r\n' 65 | 66 | 67 | def pedestrian_tracking(cfg, args): 68 | """ 69 | starts the pedestrian detection on rtsp link 70 | Args: 71 | cfg: 72 | args: 73 | 74 | Returns: 75 | 76 | """ 77 | tracker = RealTimeTracking(cfg, args) 78 | tracker.run() 79 | 80 | 81 | def trigger_process(cfg, args): 82 | """ 83 | triggers pedestrian_tracking process on rtsp link using a thread 84 | Args: 85 | cfg: 86 | args: 87 | 88 | Returns: 89 | """ 90 | try: 91 | t = Thread(target=pedestrian_tracking, args=(cfg, args)) 92 | t.start() 93 | return jsonify({"message": "Pedestrian detection started successfully"}) 94 | except Exception: 95 | return jsonify({'message': "Unexpected exception occured in process"}) 96 | 97 | 98 | @app.errorhandler(400) 99 | def bad_argument(error): 100 | return jsonify({'message': error.description['message']}) 101 | 102 | 103 | # Routes 104 | @app.route('/stream', methods=['GET']) 105 | def stream(): 106 | """ 107 | Provides video frames on http link 108 | Returns: 109 | 110 | """ 111 | return Response(gen(), 112 | mimetype='multipart/x-mixed-replace; boundary=frame') 113 | 114 | 115 | @app.route("/run", methods=['GET']) 116 | def process_manager(): 117 | """ 118 | request parameters: 119 | run (bool): 1 -> start the pedestrian tracking 120 | 0 -> stop it 121 | camera_stream: str -> rtsp link to security camera 122 | 123 | :return: 124 | """ 125 | # data = request.args 126 | data = request.args 127 | status = data['run'] 128 | status = int(status) if status.isnumeric() else abort(400, {'message': f"bad argument for run {data['run']}"}) 129 | if status == 1: 130 | # if pedestrian tracking is not running, start it off! 131 | try: 132 | if environ.get('in_progress', 'off') == 'off': 133 | global cfg, args 134 | vdo = data.get('camera_stream') 135 | if vdo is not None: 136 | args.input = int(vdo) 137 | environ['in_progress'] = 'on' 138 | return trigger_process(cfg, args) 139 | elif environ.get('in_progress') == 'on': 140 | # if pedestrian tracking is running, don't start another one (we are short of gpu resources) 141 | return jsonify({"message": " Pedestrian detection is already in progress."}) 142 | except Exception: 143 | environ['in_progress'] = 'off' 144 | return abort(503) 145 | elif status == 0: 146 | if environ.get('in_progress', 'off') == 'off': 147 | return jsonify({"message": "pedestrian detection is already terminated!"}) 148 | else: 149 | environ['in_progress'] = 'off' 150 | return jsonify({"message": "Pedestrian detection terminated!"}) 151 | 152 | 153 | if __name__ == '__main__': 154 | load_dotenv() 155 | app.config.from_object(DevelopmentConfig) 156 | 157 | # BackProcess Initialization 158 | args = parse_args() 159 | cfg = get_config() 160 | cfg.merge_from_dict(model) 161 | cfg.merge_from_dict(deep_sort_dict) 162 | # Start the flask app 163 | app.run() 164 | -------------------------------------------------------------------------------- /deep_sort/webserver/server_cfg.py: -------------------------------------------------------------------------------- 1 | """""" 2 | import sys 3 | from os.path import dirname, abspath, isfile 4 | 5 | sys.path.append(dirname(dirname(abspath(__file__)))) 6 | 7 | from dotenv import load_dotenv 8 | from utils.asserts import assert_in_env 9 | from os import getenv 10 | from os.path import join 11 | 12 | load_dotenv('.env') 13 | # Configure deep sort info 14 | deep_sort_info = dict(REID_CKPT=join(getenv('project_root'), getenv('reid_ckpt')), 15 | MAX_DIST=0.2, 16 | MIN_CONFIDENCE=.3, 17 | NMS_MAX_OVERLAP=0.5, 18 | MAX_IOU_DISTANCE=0.7, 19 | N_INIT=3, 20 | MAX_AGE=70, 21 | NN_BUDGET=100) 22 | deep_sort_dict = {'DEEPSORT': deep_sort_info} 23 | 24 | # Configure yolov3 info 25 | 26 | yolov3_info = dict(CFG=join(getenv('project_root'), getenv('yolov3_cfg')), 27 | WEIGHT=join(getenv('project_root'), getenv('yolov3_weight')), 28 | CLASS_NAMES=join(getenv('project_root'), getenv('yolov3_class_names')), 29 | SCORE_THRESH=0.5, 30 | NMS_THRESH=0.4 31 | ) 32 | yolov3_dict = {'YOLOV3': yolov3_info} 33 | 34 | # Configure yolov3-tiny info 35 | 36 | yolov3_tiny_info = dict(CFG=join(getenv('project_root'), getenv('yolov3_tiny_cfg')), 37 | WEIGHT=join(getenv('project_root'), getenv('yolov3_tiny_weight')), 38 | CLASS_NAMES=join(getenv('project_root'), getenv('yolov3_class_names')), 39 | SCORE_THRESH=0.5, 40 | NMS_THRESH=0.4 41 | ) 42 | yolov3_tiny_dict = {'YOLOV3': yolov3_tiny_info} 43 | 44 | 45 | check_list = ['project_root', 'reid_ckpt', 'yolov3_class_names', 'model_type', 'yolov3_cfg', 'yolov3_weight', 46 | 'yolov3_tiny_cfg', 'yolov3_tiny_weight', 'yolov3_class_names'] 47 | 48 | if assert_in_env(check_list): 49 | assert isfile(deep_sort_info['REID_CKPT']) 50 | if getenv('model_type') == 'yolov3': 51 | assert isfile(yolov3_info['WEIGHT']) 52 | assert isfile(yolov3_info['CFG']) 53 | assert isfile(yolov3_info['CLASS_NAMES']) 54 | model = yolov3_dict.copy() 55 | 56 | elif getenv('model_type') == 'yolov3_tiny': 57 | assert isfile(yolov3_tiny_info['WEIGHT']) 58 | assert isfile(yolov3_tiny_info['CFG']) 59 | assert isfile(yolov3_tiny_info['CLASS_NAMES']) 60 | model = yolov3_tiny_dict.copy() 61 | else: 62 | raise ValueError("Value '{}' for model_type is not valid".format(getenv('model_type'))) 63 | -------------------------------------------------------------------------------- /deep_sort/webserver/templates/index.html: -------------------------------------------------------------------------------- 1 | 2 | 3 | دوربین طبقه 2: راهرو 2 4 | 5 | 6 |

طبقه 2

7 | 8 | 9 | -------------------------------------------------------------------------------- /deep_sort/yolov3_deepsort.py: -------------------------------------------------------------------------------- 1 | import os 2 | import cv2 3 | import time 4 | import argparse 5 | import torch 6 | import warnings 7 | import numpy as np 8 | 9 | from detector import build_detector 10 | from deep_sort import build_tracker 11 | from utils.draw import draw_boxes 12 | from utils.parser import get_config 13 | from utils.log import get_logger 14 | from utils.io import write_results 15 | 16 | 17 | class VideoTracker(object): 18 | def __init__(self, cfg, args, video_path): 19 | self.cfg = cfg 20 | self.args = args 21 | self.video_path = video_path 22 | self.logger = get_logger("root") 23 | 24 | use_cuda = args.use_cuda and torch.cuda.is_available() 25 | if not use_cuda: 26 | warnings.warn("Running in cpu mode which maybe very slow!", UserWarning) 27 | 28 | if args.display: 29 | cv2.namedWindow("test", cv2.WINDOW_NORMAL) 30 | cv2.resizeWindow("test", args.display_width, args.display_height) 31 | 32 | if args.cam != -1: 33 | print("Using webcam " + str(args.cam)) 34 | self.vdo = cv2.VideoCapture(args.cam) 35 | else: 36 | self.vdo = cv2.VideoCapture() 37 | self.detector = build_detector(cfg, use_cuda=use_cuda) 38 | self.deepsort = build_tracker(cfg, use_cuda=use_cuda) 39 | self.class_names = self.detector.class_names 40 | 41 | def __enter__(self): 42 | if self.args.cam != -1: 43 | ret, frame = self.vdo.read() 44 | assert ret, "Error: Camera error" 45 | self.im_width = frame.shape[0] 46 | self.im_height = frame.shape[1] 47 | 48 | else: 49 | assert os.path.isfile(self.video_path), "Path error" 50 | self.vdo.open(self.video_path) 51 | self.im_width = int(self.vdo.get(cv2.CAP_PROP_FRAME_WIDTH)) 52 | self.im_height = int(self.vdo.get(cv2.CAP_PROP_FRAME_HEIGHT)) 53 | assert self.vdo.isOpened() 54 | 55 | if self.args.save_path: 56 | os.makedirs(self.args.save_path, exist_ok=True) 57 | 58 | # path of saved video and results 59 | self.save_video_path = os.path.join(self.args.save_path, "results.avi") 60 | self.save_results_path = os.path.join(self.args.save_path, "results.txt") 61 | 62 | # create video writer 63 | fourcc = cv2.VideoWriter_fourcc(*'MJPG') 64 | self.writer = cv2.VideoWriter(self.save_video_path, fourcc, 20, (self.im_width, self.im_height)) 65 | 66 | # logging 67 | self.logger.info("Save results to {}".format(self.args.save_path)) 68 | 69 | return self 70 | 71 | def __exit__(self, exc_type, exc_value, exc_traceback): 72 | if exc_type: 73 | print(exc_type, exc_value, exc_traceback) 74 | 75 | def run(self): 76 | results = [] 77 | idx_frame = 0 78 | while self.vdo.grab(): 79 | idx_frame += 1 80 | if idx_frame % self.args.frame_interval: 81 | continue 82 | 83 | start = time.time() 84 | _, ori_im = self.vdo.retrieve() 85 | im = cv2.cvtColor(ori_im, cv2.COLOR_BGR2RGB) 86 | 87 | # do detection 88 | bbox_xywh, cls_conf, cls_ids = self.detector(im) 89 | 90 | # select person class 91 | mask = cls_ids == 0 92 | 93 | bbox_xywh = bbox_xywh[mask] 94 | # bbox dilation just in case bbox too small, delete this line if using a better pedestrian detector 95 | bbox_xywh[:, 3:] *= 1.2 96 | cls_conf = cls_conf[mask] 97 | 98 | # do tracking 99 | outputs = self.deepsort.update(bbox_xywh, cls_conf, im) 100 | 101 | # draw boxes for visualization 102 | if len(outputs) > 0: 103 | bbox_tlwh = [] 104 | bbox_xyxy = outputs[:, :4] 105 | identities = outputs[:, -1] 106 | ori_im = draw_boxes(ori_im, bbox_xyxy, identities) 107 | 108 | for bb_xyxy in bbox_xyxy: 109 | bbox_tlwh.append(self.deepsort._xyxy_to_tlwh(bb_xyxy)) 110 | 111 | results.append((idx_frame - 1, bbox_tlwh, identities)) 112 | 113 | end = time.time() 114 | 115 | if self.args.display: 116 | cv2.imshow("test", ori_im) 117 | cv2.waitKey(1) 118 | 119 | if self.args.save_path: 120 | self.writer.write(ori_im) 121 | 122 | # save results 123 | write_results(self.save_results_path, results, 'mot') 124 | 125 | # logging 126 | self.logger.info("time: {:.03f}s, fps: {:.03f}, detection numbers: {}, tracking numbers: {}" \ 127 | .format(end - start, 1 / (end - start), bbox_xywh.shape[0], len(outputs))) 128 | 129 | 130 | def parse_args(): 131 | parser = argparse.ArgumentParser() 132 | parser.add_argument("VIDEO_PATH", type=str) 133 | parser.add_argument("--config_detection", type=str, default="./configs/yolov3.yaml") 134 | parser.add_argument("--config_deepsort", type=str, default="./configs/deep_sort.yaml") 135 | # parser.add_argument("--ignore_display", dest="display", action="store_false", default=True) 136 | parser.add_argument("--display", action="store_true") 137 | parser.add_argument("--frame_interval", type=int, default=1) 138 | parser.add_argument("--display_width", type=int, default=800) 139 | parser.add_argument("--display_height", type=int, default=600) 140 | parser.add_argument("--save_path", type=str, default="./output/") 141 | parser.add_argument("--cpu", dest="use_cuda", action="store_false", default=True) 142 | parser.add_argument("--camera", action="store", dest="cam", type=int, default="-1") 143 | return parser.parse_args() 144 | 145 | 146 | if __name__ == "__main__": 147 | args = parse_args() 148 | cfg = get_config() 149 | cfg.merge_from_file(args.config_detection) 150 | cfg.merge_from_file(args.config_deepsort) 151 | 152 | with VideoTracker(cfg, args, video_path=args.VIDEO_PATH) as vdo_trk: 153 | vdo_trk.run() 154 | -------------------------------------------------------------------------------- /deep_sort/yolov3_deepsort_eval.py: -------------------------------------------------------------------------------- 1 | import os 2 | import os.path as osp 3 | import logging 4 | import argparse 5 | from pathlib import Path 6 | 7 | from utils.log import get_logger 8 | from yolov3_deepsort import VideoTracker 9 | from utils.parser import get_config 10 | 11 | import motmetrics as mm 12 | mm.lap.default_solver = 'lap' 13 | from utils.evaluation import Evaluator 14 | 15 | def mkdir_if_missing(dir): 16 | os.makedirs(dir, exist_ok=True) 17 | 18 | def main(data_root='', seqs=('',), args=""): 19 | logger = get_logger() 20 | logger.setLevel(logging.INFO) 21 | data_type = 'mot' 22 | result_root = os.path.join(Path(data_root), "mot_results") 23 | mkdir_if_missing(result_root) 24 | 25 | cfg = get_config() 26 | cfg.merge_from_file(args.config_detection) 27 | cfg.merge_from_file(args.config_deepsort) 28 | 29 | # run tracking 30 | accs = [] 31 | for seq in seqs: 32 | logger.info('start seq: {}'.format(seq)) 33 | result_filename = os.path.join(result_root, '{}.txt'.format(seq)) 34 | video_path = data_root+"/"+seq+"/video/video.mp4" 35 | 36 | with VideoTracker(cfg, args, video_path, result_filename) as vdo_trk: 37 | vdo_trk.run() 38 | 39 | # eval 40 | logger.info('Evaluate seq: {}'.format(seq)) 41 | evaluator = Evaluator(data_root, seq, data_type) 42 | accs.append(evaluator.eval_file(result_filename)) 43 | 44 | # get summary 45 | metrics = mm.metrics.motchallenge_metrics 46 | mh = mm.metrics.create() 47 | summary = Evaluator.get_summary(accs, seqs, metrics) 48 | strsummary = mm.io.render_summary( 49 | summary, 50 | formatters=mh.formatters, 51 | namemap=mm.io.motchallenge_metric_names 52 | ) 53 | print(strsummary) 54 | Evaluator.save_summary(summary, os.path.join(result_root, 'summary_global.xlsx')) 55 | 56 | 57 | def parse_args(): 58 | parser = argparse.ArgumentParser() 59 | parser.add_argument("--config_detection", type=str, default="./configs/yolov3.yaml") 60 | parser.add_argument("--config_deepsort", type=str, default="./configs/deep_sort.yaml") 61 | parser.add_argument("--ignore_display", dest="display", action="store_false", default=False) 62 | parser.add_argument("--frame_interval", type=int, default=1) 63 | parser.add_argument("--display_width", type=int, default=800) 64 | parser.add_argument("--display_height", type=int, default=600) 65 | parser.add_argument("--save_path", type=str, default="./demo/demo.avi") 66 | parser.add_argument("--cpu", dest="use_cuda", action="store_false", default=True) 67 | parser.add_argument("--camera", action="store", dest="cam", type=int, default="-1") 68 | return parser.parse_args() 69 | 70 | if __name__ == '__main__': 71 | args = parse_args() 72 | 73 | seqs_str = '''MOT16-02 74 | MOT16-04 75 | MOT16-05 76 | MOT16-09 77 | MOT16-10 78 | MOT16-11 79 | MOT16-13 80 | ''' 81 | data_root = 'data/dataset/MOT16/train/' 82 | 83 | seqs = [seq.strip() for seq in seqs_str.split()] 84 | 85 | main(data_root=data_root, 86 | seqs=seqs, 87 | args=args) -------------------------------------------------------------------------------- /requirements.txt: -------------------------------------------------------------------------------- 1 | # pip install -U -r requirements.txt 2 | Cython 3 | matplotlib>=3.2.2 4 | numpy>=1.18.5 5 | opencv-python>=4.1.2 6 | pillow 7 | easydict 8 | pycocotools>=2.0 9 | PyYAML>=5.3 10 | scipy>=1.4.1 11 | tensorboard>=2.2 12 | tqdm>=4.41.0 13 | -------------------------------------------------------------------------------- /templates/index.html: -------------------------------------------------------------------------------- 1 | 3 | 4 | 5 | 6 | 多目标跟踪 7 | 8 | 9 |

多目标跟踪

10 | 11 |
12 | 13 |
14 | 15 | -------------------------------------------------------------------------------- /yolov5/.dockerignore: -------------------------------------------------------------------------------- 1 | # Repo-specific DockerIgnore ------------------------------------------------------------------------------------------- 2 | # .git 3 | .cache 4 | .idea 5 | runs 6 | output 7 | coco 8 | storage.googleapis.com 9 | 10 | data/samples/* 11 | **/results*.txt 12 | *.jpg 13 | 14 | # Neural Network weights ----------------------------------------------------------------------------------------------- 15 | **/*.weights 16 | **/*.pt 17 | **/*.pth 18 | **/*.onnx 19 | **/*.mlmodel 20 | **/*.torchscript 21 | 22 | 23 | # Below Copied From .gitignore ----------------------------------------------------------------------------------------- 24 | # Below Copied From .gitignore ----------------------------------------------------------------------------------------- 25 | 26 | 27 | # GitHub Python GitIgnore ---------------------------------------------------------------------------------------------- 28 | # Byte-compiled / optimized / DLL files 29 | __pycache__/ 30 | *.py[cod] 31 | *$py.class 32 | 33 | # C extensions 34 | *.so 35 | 36 | # Distribution / packaging 37 | .Python 38 | env/ 39 | build/ 40 | develop-eggs/ 41 | dist/ 42 | downloads/ 43 | eggs/ 44 | .eggs/ 45 | lib/ 46 | lib64/ 47 | parts/ 48 | sdist/ 49 | var/ 50 | wheels/ 51 | *.egg-info/ 52 | .installed.cfg 53 | *.egg 54 | 55 | # PyInstaller 56 | # Usually these files are written by a python script from a template 57 | # before PyInstaller builds the exe, so as to inject date/other infos into it. 58 | *.manifest 59 | *.spec 60 | 61 | # Installer logs 62 | pip-log.txt 63 | pip-delete-this-directory.txt 64 | 65 | # Unit test / coverage reports 66 | htmlcov/ 67 | .tox/ 68 | .coverage 69 | .coverage.* 70 | .cache 71 | nosetests.xml 72 | coverage.xml 73 | *.cover 74 | .hypothesis/ 75 | 76 | # Translations 77 | *.mo 78 | *.pot 79 | 80 | # Django stuff: 81 | *.log 82 | local_settings.py 83 | 84 | # Flask stuff: 85 | instance/ 86 | .webassets-cache 87 | 88 | # Scrapy stuff: 89 | .scrapy 90 | 91 | # Sphinx documentation 92 | docs/_build/ 93 | 94 | # PyBuilder 95 | target/ 96 | 97 | # Jupyter Notebook 98 | .ipynb_checkpoints 99 | 100 | # pyenv 101 | .python-version 102 | 103 | # celery beat schedule file 104 | celerybeat-schedule 105 | 106 | # SageMath parsed files 107 | *.sage.py 108 | 109 | # dotenv 110 | .env 111 | 112 | # virtualenv 113 | .venv 114 | venv*/ 115 | ENV/ 116 | 117 | # Spyder project settings 118 | .spyderproject 119 | .spyproject 120 | 121 | # Rope project settings 122 | .ropeproject 123 | 124 | # mkdocs documentation 125 | /site 126 | 127 | # mypy 128 | .mypy_cache/ 129 | 130 | 131 | # https://github.com/github/gitignore/blob/master/Global/macOS.gitignore ----------------------------------------------- 132 | 133 | # General 134 | .DS_Store 135 | .AppleDouble 136 | .LSOverride 137 | 138 | # Icon must end with two \r 139 | Icon 140 | Icon? 141 | 142 | # Thumbnails 143 | ._* 144 | 145 | # Files that might appear in the root of a volume 146 | .DocumentRevisions-V100 147 | .fseventsd 148 | .Spotlight-V100 149 | .TemporaryItems 150 | .Trashes 151 | .VolumeIcon.icns 152 | .com.apple.timemachine.donotpresent 153 | 154 | # Directories potentially created on remote AFP share 155 | .AppleDB 156 | .AppleDesktop 157 | Network Trash Folder 158 | Temporary Items 159 | .apdisk 160 | 161 | 162 | # https://github.com/github/gitignore/blob/master/Global/JetBrains.gitignore 163 | # Covers JetBrains IDEs: IntelliJ, RubyMine, PhpStorm, AppCode, PyCharm, CLion, Android Studio and WebStorm 164 | # Reference: https://intellij-support.jetbrains.com/hc/en-us/articles/206544839 165 | 166 | # User-specific stuff: 167 | .idea/* 168 | .idea/**/workspace.xml 169 | .idea/**/tasks.xml 170 | .idea/dictionaries 171 | .html # Bokeh Plots 172 | .pg # TensorFlow Frozen Graphs 173 | .avi # videos 174 | 175 | # Sensitive or high-churn files: 176 | .idea/**/dataSources/ 177 | .idea/**/dataSources.ids 178 | .idea/**/dataSources.local.xml 179 | .idea/**/sqlDataSources.xml 180 | .idea/**/dynamic.xml 181 | .idea/**/uiDesigner.xml 182 | 183 | # Gradle: 184 | .idea/**/gradle.xml 185 | .idea/**/libraries 186 | 187 | # CMake 188 | cmake-build-debug/ 189 | cmake-build-release/ 190 | 191 | # Mongo Explorer plugin: 192 | .idea/**/mongoSettings.xml 193 | 194 | ## File-based project format: 195 | *.iws 196 | 197 | ## Plugin-specific files: 198 | 199 | # IntelliJ 200 | out/ 201 | 202 | # mpeltonen/sbt-idea plugin 203 | .idea_modules/ 204 | 205 | # JIRA plugin 206 | atlassian-ide-plugin.xml 207 | 208 | # Cursive Clojure plugin 209 | .idea/replstate.xml 210 | 211 | # Crashlytics plugin (for Android Studio and IntelliJ) 212 | com_crashlytics_export_strings.xml 213 | crashlytics.properties 214 | crashlytics-build.properties 215 | fabric.properties 216 | -------------------------------------------------------------------------------- /yolov5/.gitattributes: -------------------------------------------------------------------------------- 1 | # this drop notebooks from GitHub language stats 2 | *.ipynb linguist-vendored 3 | -------------------------------------------------------------------------------- /yolov5/.github/ISSUE_TEMPLATE/--bug-report.md: -------------------------------------------------------------------------------- 1 | --- 2 | name: "\U0001F41BBug report" 3 | about: Create a report to help us improve 4 | title: '' 5 | labels: bug 6 | assignees: '' 7 | 8 | --- 9 | 10 | Before submitting a bug report, please be aware that your issue **must be reproducible** with all of the following, otherwise it is non-actionable, and we can not help you: 11 | - **Current repo**: run `git fetch && git status -uno` to check and `git pull` to update repo 12 | - **Common dataset**: coco.yaml or coco128.yaml 13 | - **Common environment**: Colab, Google Cloud, or Docker image. See https://github.com/ultralytics/yolov5#environments 14 | 15 | If this is a custom dataset/training question you **must include** your `train*.jpg`, `test*.jpg` and `results.png` figures, or we can not help you. You can generate these with `utils.plot_results()`. 16 | 17 | 18 | ## 🐛 Bug 19 | A clear and concise description of what the bug is. 20 | 21 | 22 | ## To Reproduce (REQUIRED) 23 | 24 | Input: 25 | ``` 26 | import torch 27 | 28 | a = torch.tensor([5]) 29 | c = a / 0 30 | ``` 31 | 32 | Output: 33 | ``` 34 | Traceback (most recent call last): 35 | File "/Users/glennjocher/opt/anaconda3/envs/env1/lib/python3.7/site-packages/IPython/core/interactiveshell.py", line 3331, in run_code 36 | exec(code_obj, self.user_global_ns, self.user_ns) 37 | File "", line 5, in 38 | c = a / 0 39 | RuntimeError: ZeroDivisionError 40 | ``` 41 | 42 | 43 | ## Expected behavior 44 | A clear and concise description of what you expected to happen. 45 | 46 | 47 | ## Environment 48 | If applicable, add screenshots to help explain your problem. 49 | 50 | - OS: [e.g. Ubuntu] 51 | - GPU [e.g. 2080 Ti] 52 | 53 | 54 | ## Additional context 55 | Add any other context about the problem here. 56 | -------------------------------------------------------------------------------- /yolov5/.github/ISSUE_TEMPLATE/--feature-request.md: -------------------------------------------------------------------------------- 1 | --- 2 | name: "\U0001F680Feature request" 3 | about: Suggest an idea for this project 4 | title: '' 5 | labels: enhancement 6 | assignees: '' 7 | 8 | --- 9 | 10 | ## 🚀 Feature 11 | 12 | 13 | ## Motivation 14 | 15 | 16 | 17 | ## Pitch 18 | 19 | 20 | 21 | ## Alternatives 22 | 23 | 24 | 25 | ## Additional context 26 | 27 | 28 | -------------------------------------------------------------------------------- /yolov5/.github/ISSUE_TEMPLATE/-question.md: -------------------------------------------------------------------------------- 1 | --- 2 | name: "❓Question" 3 | about: Ask a general question 4 | title: '' 5 | labels: question 6 | assignees: '' 7 | 8 | --- 9 | 10 | ## ❔Question 11 | 12 | 13 | ## Additional context 14 | -------------------------------------------------------------------------------- /yolov5/.github/workflows/ci-testing.yml: -------------------------------------------------------------------------------- 1 | name: CI CPU testing 2 | 3 | on: # https://help.github.com/en/actions/reference/events-that-trigger-workflows 4 | push: 5 | pull_request: 6 | schedule: 7 | - cron: "0 0 * * *" 8 | 9 | jobs: 10 | cpu-tests: 11 | 12 | runs-on: ${{ matrix.os }} 13 | strategy: 14 | fail-fast: false 15 | matrix: 16 | os: [ubuntu-latest, macos-latest, windows-latest] 17 | python-version: [3.8] 18 | model: ['yolov5s'] # models to test 19 | 20 | # Timeout: https://stackoverflow.com/a/59076067/4521646 21 | timeout-minutes: 50 22 | steps: 23 | - uses: actions/checkout@v2 24 | - name: Set up Python ${{ matrix.python-version }} 25 | uses: actions/setup-python@v2 26 | with: 27 | python-version: ${{ matrix.python-version }} 28 | 29 | # Note: This uses an internal pip API and may not always work 30 | # https://github.com/actions/cache/blob/master/examples.md#multiple-oss-in-a-workflow 31 | - name: Get pip cache 32 | id: pip-cache 33 | run: | 34 | python -c "from pip._internal.locations import USER_CACHE_DIR; print('::set-output name=dir::' + USER_CACHE_DIR)" 35 | 36 | - name: Cache pip 37 | uses: actions/cache@v1 38 | with: 39 | path: ${{ steps.pip-cache.outputs.dir }} 40 | key: ${{ runner.os }}-${{ matrix.python-version }}-pip-${{ hashFiles('requirements.txt') }} 41 | restore-keys: | 42 | ${{ runner.os }}-${{ matrix.python-version }}-pip- 43 | 44 | - name: Install dependencies 45 | run: | 46 | python -m pip install --upgrade pip 47 | pip install -qr requirements.txt -f https://download.pytorch.org/whl/cpu/torch_stable.html 48 | pip install -q onnx 49 | python --version 50 | pip --version 51 | pip list 52 | shell: bash 53 | 54 | - name: Download data 55 | run: | 56 | curl -L -o temp.zip https://github.com/ultralytics/yolov5/releases/download/v1.0/coco128.zip 57 | unzip -q temp.zip -d ../ 58 | rm temp.zip 59 | 60 | - name: Tests workflow 61 | run: | 62 | export PYTHONPATH="$PWD" # to run *.py. files in subdirectories 63 | di=cpu # inference devices # define device 64 | 65 | # train 66 | python train.py --img 256 --batch 8 --weights weights/${{ matrix.model }}.pt --cfg models/${{ matrix.model }}.yaml --epochs 1 --device $di 67 | # detect 68 | python detect.py --weights weights/${{ matrix.model }}.pt --device $di 69 | python detect.py --weights runs/exp0/weights/last.pt --device $di 70 | # test 71 | python test.py --img 256 --batch 8 --weights weights/${{ matrix.model }}.pt --device $di 72 | python test.py --img 256 --batch 8 --weights runs/exp0/weights/last.pt --device $di 73 | 74 | python models/yolo.py --cfg models/${{ matrix.model }}.yaml # inspect 75 | python models/export.py --img 256 --batch 1 --weights weights/${{ matrix.model }}.pt # export 76 | shell: bash 77 | -------------------------------------------------------------------------------- /yolov5/.github/workflows/greetings.yml: -------------------------------------------------------------------------------- 1 | name: Greetings 2 | 3 | on: [pull_request_target, issues] 4 | 5 | jobs: 6 | greeting: 7 | runs-on: ubuntu-latest 8 | steps: 9 | - uses: actions/first-interaction@v1 10 | with: 11 | repo-token: ${{ secrets.GITHUB_TOKEN }} 12 | pr-message: | 13 | Hello @${{ github.actor }}, thank you for submitting a PR! To allow your work to be integrated as seamlessly as possible, we advise you to: 14 | - Verify your PR is **up-to-date with origin/master.** If your PR is behind origin/master update by running the following, replacing 'feature' with the name of your local branch: 15 | ```bash 16 | git remote add upstream https://github.com/ultralytics/yolov5.git 17 | git fetch upstream 18 | git checkout feature # <----- replace 'feature' with local branch name 19 | git rebase upstream/master 20 | git push -u origin -f 21 | ``` 22 | - Verify all Continuous Integration (CI) **checks are passing**. 23 | - Reduce changes to the absolute **minimum** required for your bug fix or feature addition. _"It is not daily increase but daily decrease, hack away the unessential. The closer to the source, the less wastage there is."_ -Bruce Lee 24 | 25 | issue-message: | 26 | Hello @${{ github.actor }}, thank you for your interest in our work! Please visit our [Custom Training Tutorial](https://github.com/ultralytics/yolov5/wiki/Train-Custom-Data) to get started, and see our [Jupyter Notebook](https://github.com/ultralytics/yolov5/blob/master/tutorial.ipynb) Open In Colab, [Docker Image](https://hub.docker.com/r/ultralytics/yolov5), and [Google Cloud Quickstart Guide](https://github.com/ultralytics/yolov5/wiki/GCP-Quickstart) for example environments. 27 | 28 | If this is a bug report, please provide screenshots and **minimum viable code to reproduce your issue**, otherwise we can not help you. 29 | 30 | If this is a custom model or data training question, please note Ultralytics does **not** provide free personal support. As a leader in vision ML and AI, we do offer professional consulting, from simple expert advice up to delivery of fully customized, end-to-end production solutions for our clients, such as: 31 | - **Cloud-based AI** systems operating on **hundreds of HD video streams in realtime.** 32 | - **Edge AI** integrated into custom iOS and Android apps for realtime **30 FPS video inference.** 33 | - **Custom data training**, hyperparameter evolution, and model exportation to any destination. 34 | 35 | For more information please visit https://www.ultralytics.com. 36 | -------------------------------------------------------------------------------- /yolov5/.github/workflows/rebase.yml: -------------------------------------------------------------------------------- 1 | name: Automatic Rebase 2 | # https://github.com/marketplace/actions/automatic-rebase 3 | 4 | on: 5 | issue_comment: 6 | types: [created] 7 | 8 | jobs: 9 | rebase: 10 | name: Rebase 11 | if: github.event.issue.pull_request != '' && contains(github.event.comment.body, '/rebase') 12 | runs-on: ubuntu-latest 13 | steps: 14 | - name: Checkout the latest code 15 | uses: actions/checkout@v2 16 | with: 17 | fetch-depth: 0 18 | - name: Automatic Rebase 19 | uses: cirrus-actions/rebase@1.3.1 20 | env: 21 | GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }} 22 | -------------------------------------------------------------------------------- /yolov5/.github/workflows/stale.yml: -------------------------------------------------------------------------------- 1 | name: Close stale issues 2 | on: 3 | schedule: 4 | - cron: "0 0 * * *" 5 | 6 | jobs: 7 | stale: 8 | runs-on: ubuntu-latest 9 | steps: 10 | - uses: actions/stale@v1 11 | with: 12 | repo-token: ${{ secrets.GITHUB_TOKEN }} 13 | stale-issue-message: 'This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.' 14 | stale-pr-message: 'This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.' 15 | days-before-stale: 30 16 | days-before-close: 5 17 | exempt-issue-label: 'documentation,tutorial' 18 | operations-per-run: 100 # The maximum number of operations per run, used to control rate limiting. 19 | -------------------------------------------------------------------------------- /yolov5/.gitignore: -------------------------------------------------------------------------------- 1 | # Repo-specific GitIgnore ---------------------------------------------------------------------------------------------- 2 | *.jpg 3 | *.jpeg 4 | *.png 5 | *.bmp 6 | *.tif 7 | *.tiff 8 | *.heic 9 | *.JPG 10 | *.JPEG 11 | *.PNG 12 | *.BMP 13 | *.TIF 14 | *.TIFF 15 | *.HEIC 16 | *.mp4 17 | *.mov 18 | *.MOV 19 | *.avi 20 | *.data 21 | *.json 22 | 23 | *.cfg 24 | !cfg/yolov3*.cfg 25 | 26 | storage.googleapis.com 27 | runs/* 28 | data/* 29 | !data/samples/zidane.jpg 30 | !data/samples/bus.jpg 31 | !data/coco.names 32 | !data/coco_paper.names 33 | !data/coco.data 34 | !data/coco_*.data 35 | !data/coco_*.txt 36 | !data/trainvalno5k.shapes 37 | !data/*.sh 38 | 39 | pycocotools/* 40 | results*.txt 41 | gcp_test*.sh 42 | 43 | # MATLAB GitIgnore ----------------------------------------------------------------------------------------------------- 44 | *.m~ 45 | *.mat 46 | !targets*.mat 47 | 48 | # Neural Network weights ----------------------------------------------------------------------------------------------- 49 | *.weights 50 | *.pt 51 | *.onnx 52 | *.mlmodel 53 | *.torchscript 54 | darknet53.conv.74 55 | yolov3-tiny.conv.15 56 | 57 | # GitHub Python GitIgnore ---------------------------------------------------------------------------------------------- 58 | # Byte-compiled / optimized / DLL files 59 | __pycache__/ 60 | *.py[cod] 61 | *$py.class 62 | 63 | # C extensions 64 | *.so 65 | 66 | # Distribution / packaging 67 | .Python 68 | env/ 69 | build/ 70 | develop-eggs/ 71 | dist/ 72 | downloads/ 73 | eggs/ 74 | .eggs/ 75 | lib/ 76 | lib64/ 77 | parts/ 78 | sdist/ 79 | var/ 80 | wheels/ 81 | *.egg-info/ 82 | .installed.cfg 83 | *.egg 84 | 85 | # PyInstaller 86 | # Usually these files are written by a python script from a template 87 | # before PyInstaller builds the exe, so as to inject date/other infos into it. 88 | *.manifest 89 | *.spec 90 | 91 | # Installer logs 92 | pip-log.txt 93 | pip-delete-this-directory.txt 94 | 95 | # Unit test / coverage reports 96 | htmlcov/ 97 | .tox/ 98 | .coverage 99 | .coverage.* 100 | .cache 101 | nosetests.xml 102 | coverage.xml 103 | *.cover 104 | .hypothesis/ 105 | 106 | # Translations 107 | *.mo 108 | *.pot 109 | 110 | # Django stuff: 111 | *.log 112 | local_settings.py 113 | 114 | # Flask stuff: 115 | instance/ 116 | .webassets-cache 117 | 118 | # Scrapy stuff: 119 | .scrapy 120 | 121 | # Sphinx documentation 122 | docs/_build/ 123 | 124 | # PyBuilder 125 | target/ 126 | 127 | # Jupyter Notebook 128 | .ipynb_checkpoints 129 | 130 | # pyenv 131 | .python-version 132 | 133 | # celery beat schedule file 134 | celerybeat-schedule 135 | 136 | # SageMath parsed files 137 | *.sage.py 138 | 139 | # dotenv 140 | .env 141 | 142 | # virtualenv 143 | .venv 144 | venv/ 145 | ENV/ 146 | 147 | # Spyder project settings 148 | .spyderproject 149 | .spyproject 150 | 151 | # Rope project settings 152 | .ropeproject 153 | 154 | # mkdocs documentation 155 | /site 156 | 157 | # mypy 158 | .mypy_cache/ 159 | 160 | 161 | # https://github.com/github/gitignore/blob/master/Global/macOS.gitignore ----------------------------------------------- 162 | 163 | # General 164 | .DS_Store 165 | .AppleDouble 166 | .LSOverride 167 | 168 | # Icon must end with two \r 169 | Icon 170 | Icon? 171 | 172 | # Thumbnails 173 | ._* 174 | 175 | # Files that might appear in the root of a volume 176 | .DocumentRevisions-V100 177 | .fseventsd 178 | .Spotlight-V100 179 | .TemporaryItems 180 | .Trashes 181 | .VolumeIcon.icns 182 | .com.apple.timemachine.donotpresent 183 | 184 | # Directories potentially created on remote AFP share 185 | .AppleDB 186 | .AppleDesktop 187 | Network Trash Folder 188 | Temporary Items 189 | .apdisk 190 | 191 | 192 | # https://github.com/github/gitignore/blob/master/Global/JetBrains.gitignore 193 | # Covers JetBrains IDEs: IntelliJ, RubyMine, PhpStorm, AppCode, PyCharm, CLion, Android Studio and WebStorm 194 | # Reference: https://intellij-support.jetbrains.com/hc/en-us/articles/206544839 195 | 196 | # User-specific stuff: 197 | .idea/* 198 | .idea/**/workspace.xml 199 | .idea/**/tasks.xml 200 | .idea/dictionaries 201 | .html # Bokeh Plots 202 | .pg # TensorFlow Frozen Graphs 203 | .avi # videos 204 | 205 | # Sensitive or high-churn files: 206 | .idea/**/dataSources/ 207 | .idea/**/dataSources.ids 208 | .idea/**/dataSources.local.xml 209 | .idea/**/sqlDataSources.xml 210 | .idea/**/dynamic.xml 211 | .idea/**/uiDesigner.xml 212 | 213 | # Gradle: 214 | .idea/**/gradle.xml 215 | .idea/**/libraries 216 | 217 | # CMake 218 | cmake-build-debug/ 219 | cmake-build-release/ 220 | 221 | # Mongo Explorer plugin: 222 | .idea/**/mongoSettings.xml 223 | 224 | ## File-based project format: 225 | *.iws 226 | 227 | ## Plugin-specific files: 228 | 229 | # IntelliJ 230 | out/ 231 | 232 | # mpeltonen/sbt-idea plugin 233 | .idea_modules/ 234 | 235 | # JIRA plugin 236 | atlassian-ide-plugin.xml 237 | 238 | # Cursive Clojure plugin 239 | .idea/replstate.xml 240 | 241 | # Crashlytics plugin (for Android Studio and IntelliJ) 242 | com_crashlytics_export_strings.xml 243 | crashlytics.properties 244 | crashlytics-build.properties 245 | fabric.properties 246 | -------------------------------------------------------------------------------- /yolov5/Dockerfile: -------------------------------------------------------------------------------- 1 | # Start FROM Nvidia PyTorch image https://ngc.nvidia.com/catalog/containers/nvidia:pytorch 2 | FROM nvcr.io/nvidia/pytorch:20.08-py3 3 | 4 | # Install dependencies 5 | # COPY requirements.txt . 6 | # RUN pip install -r requirements.txt 7 | RUN pip install gsutil 8 | 9 | # Create working directory 10 | RUN mkdir -p /usr/src/app 11 | WORKDIR /usr/src/app 12 | 13 | # Copy contents 14 | COPY . /usr/src/app 15 | 16 | # Copy weights 17 | #RUN python3 -c "from models import *; \ 18 | #attempt_download('weights/yolov5s.pt'); \ 19 | #attempt_download('weights/yolov5m.pt'); \ 20 | #attempt_download('weights/yolov5l.pt')" 21 | 22 | 23 | # --------------------------------------------------- Extras Below --------------------------------------------------- 24 | 25 | # Build and Push 26 | # t=ultralytics/yolov5:latest && sudo docker build -t $t . && sudo docker push $t 27 | # for v in {300..303}; do t=ultralytics/coco:v$v && sudo docker build -t $t . && sudo docker push $t; done 28 | 29 | # Pull and Run 30 | # t=ultralytics/yolov5:latest && sudo docker pull $t && sudo docker run -it --ipc=host $t 31 | 32 | # Pull and Run with local directory access 33 | # t=ultralytics/yolov5:latest && sudo docker pull $t && sudo docker run -it --ipc=host --gpus all -v "$(pwd)"/coco:/usr/src/coco $t 34 | 35 | # Kill all 36 | # sudo docker kill "$(sudo docker ps -q)" 37 | 38 | # Kill all image-based 39 | # sudo docker kill $(sudo docker ps -a -q --filter ancestor=ultralytics/yolov5:latest) 40 | 41 | # Bash into running container 42 | # sudo docker container exec -it ba65811811ab bash 43 | 44 | # Bash into stopped container 45 | # sudo docker commit 092b16b25c5b usr/resume && sudo docker run -it --gpus all --ipc=host -v "$(pwd)"/coco:/usr/src/coco --entrypoint=sh usr/resume 46 | 47 | # Send weights to GCP 48 | # python -c "from utils.general import *; strip_optimizer('runs/exp0_*/weights/best.pt', 'tmp.pt')" && gsutil cp tmp.pt gs://*.pt 49 | 50 | # Clean up 51 | # docker system prune -a --volumes 52 | -------------------------------------------------------------------------------- /yolov5/hubconf.py: -------------------------------------------------------------------------------- 1 | """File for accessing YOLOv5 via PyTorch Hub https://pytorch.org/hub/ 2 | 3 | Usage: 4 | import torch 5 | model = torch.hub.load('ultralytics/yolov5', 'yolov5s', pretrained=True, channels=3, classes=80) 6 | """ 7 | 8 | dependencies = ['torch', 'yaml'] 9 | import os 10 | 11 | import torch 12 | 13 | from models.yolo import Model 14 | from utils.google_utils import attempt_download 15 | 16 | 17 | def create(name, pretrained, channels, classes): 18 | """Creates a specified YOLOv5 model 19 | 20 | Arguments: 21 | name (str): name of model, i.e. 'yolov5s' 22 | pretrained (bool): load pretrained weights into the model 23 | channels (int): number of input channels 24 | classes (int): number of model classes 25 | 26 | Returns: 27 | pytorch model 28 | """ 29 | config = os.path.join(os.path.dirname(__file__), 'models', '%s.yaml' % name) # model.yaml path 30 | try: 31 | model = Model(config, channels, classes) 32 | if pretrained: 33 | ckpt = '%s.pt' % name # checkpoint filename 34 | attempt_download(ckpt) # download if not found locally 35 | state_dict = torch.load(ckpt, map_location=torch.device('cpu'))['model'].float().state_dict() # to FP32 36 | state_dict = {k: v for k, v in state_dict.items() if model.state_dict()[k].shape == v.shape} # filter 37 | model.load_state_dict(state_dict, strict=False) # load 38 | return model 39 | 40 | except Exception as e: 41 | help_url = 'https://github.com/ultralytics/yolov5/issues/36' 42 | s = 'Cache maybe be out of date, deleting cache and retrying may solve this. See %s for help.' % help_url 43 | raise Exception(s) from e 44 | 45 | 46 | def yolov5s(pretrained=False, channels=3, classes=80): 47 | """YOLOv5-small model from https://github.com/ultralytics/yolov5 48 | 49 | Arguments: 50 | pretrained (bool): load pretrained weights into the model, default=False 51 | channels (int): number of input channels, default=3 52 | classes (int): number of model classes, default=80 53 | 54 | Returns: 55 | pytorch model 56 | """ 57 | return create('yolov5s', pretrained, channels, classes) 58 | 59 | 60 | def yolov5m(pretrained=False, channels=3, classes=80): 61 | """YOLOv5-medium model from https://github.com/ultralytics/yolov5 62 | 63 | Arguments: 64 | pretrained (bool): load pretrained weights into the model, default=False 65 | channels (int): number of input channels, default=3 66 | classes (int): number of model classes, default=80 67 | 68 | Returns: 69 | pytorch model 70 | """ 71 | return create('yolov5m', pretrained, channels, classes) 72 | 73 | 74 | def yolov5l(pretrained=False, channels=3, classes=80): 75 | """YOLOv5-large model from https://github.com/ultralytics/yolov5 76 | 77 | Arguments: 78 | pretrained (bool): load pretrained weights into the model, default=False 79 | channels (int): number of input channels, default=3 80 | classes (int): number of model classes, default=80 81 | 82 | Returns: 83 | pytorch model 84 | """ 85 | return create('yolov5l', pretrained, channels, classes) 86 | 87 | 88 | def yolov5x(pretrained=False, channels=3, classes=80): 89 | """YOLOv5-xlarge model from https://github.com/ultralytics/yolov5 90 | 91 | Arguments: 92 | pretrained (bool): load pretrained weights into the model, default=False 93 | channels (int): number of input channels, default=3 94 | classes (int): number of model classes, default=80 95 | 96 | Returns: 97 | pytorch model 98 | """ 99 | return create('yolov5x', pretrained, channels, classes) 100 | -------------------------------------------------------------------------------- /yolov5/models/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/yolov5/models/__init__.py -------------------------------------------------------------------------------- /yolov5/models/common.py: -------------------------------------------------------------------------------- 1 | # This file contains modules common to various models 2 | import math 3 | 4 | import torch 5 | import torch.nn as nn 6 | 7 | 8 | def autopad(k, p=None): # kernel, padding 9 | # Pad to 'same' 10 | if p is None: 11 | p = k // 2 if isinstance(k, int) else [x // 2 for x in k] # auto-pad 12 | return p 13 | 14 | 15 | def DWConv(c1, c2, k=1, s=1, act=True): 16 | # Depthwise convolution 17 | return Conv(c1, c2, k, s, g=math.gcd(c1, c2), act=act) 18 | 19 | 20 | class Conv(nn.Module): 21 | # Standard convolution 22 | def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True): # ch_in, ch_out, kernel, stride, padding, groups 23 | super(Conv, self).__init__() 24 | self.conv = nn.Conv2d(c1, c2, k, s, autopad(k, p), groups=g, bias=False) 25 | self.bn = nn.BatchNorm2d(c2) 26 | self.act = nn.Hardswish() if act else nn.Identity() 27 | 28 | def forward(self, x): 29 | return self.act(self.bn(self.conv(x))) 30 | 31 | def fuseforward(self, x): 32 | return self.act(self.conv(x)) 33 | 34 | 35 | class Bottleneck(nn.Module): 36 | # Standard bottleneck 37 | def __init__(self, c1, c2, shortcut=True, g=1, e=0.5): # ch_in, ch_out, shortcut, groups, expansion 38 | super(Bottleneck, self).__init__() 39 | c_ = int(c2 * e) # hidden channels 40 | self.cv1 = Conv(c1, c_, 1, 1) 41 | self.cv2 = Conv(c_, c2, 3, 1, g=g) 42 | self.add = shortcut and c1 == c2 43 | 44 | def forward(self, x): 45 | return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x)) 46 | 47 | 48 | class BottleneckCSP(nn.Module): 49 | # CSP Bottleneck https://github.com/WongKinYiu/CrossStagePartialNetworks 50 | def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5): # ch_in, ch_out, number, shortcut, groups, expansion 51 | super(BottleneckCSP, self).__init__() 52 | c_ = int(c2 * e) # hidden channels 53 | self.cv1 = Conv(c1, c_, 1, 1) 54 | self.cv2 = nn.Conv2d(c1, c_, 1, 1, bias=False) 55 | self.cv3 = nn.Conv2d(c_, c_, 1, 1, bias=False) 56 | self.cv4 = Conv(2 * c_, c2, 1, 1) 57 | self.bn = nn.BatchNorm2d(2 * c_) # applied to cat(cv2, cv3) 58 | self.act = nn.LeakyReLU(0.1, inplace=True) 59 | self.m = nn.Sequential(*[Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)]) 60 | 61 | def forward(self, x): 62 | y1 = self.cv3(self.m(self.cv1(x))) 63 | y2 = self.cv2(x) 64 | return self.cv4(self.act(self.bn(torch.cat((y1, y2), dim=1)))) 65 | 66 | 67 | class SPP(nn.Module): 68 | # Spatial pyramid pooling layer used in YOLOv3-SPP 69 | def __init__(self, c1, c2, k=(5, 9, 13)): 70 | super(SPP, self).__init__() 71 | c_ = c1 // 2 # hidden channels 72 | self.cv1 = Conv(c1, c_, 1, 1) 73 | self.cv2 = Conv(c_ * (len(k) + 1), c2, 1, 1) 74 | self.m = nn.ModuleList([nn.MaxPool2d(kernel_size=x, stride=1, padding=x // 2) for x in k]) 75 | 76 | def forward(self, x): 77 | x = self.cv1(x) 78 | return self.cv2(torch.cat([x] + [m(x) for m in self.m], 1)) 79 | 80 | 81 | class Focus(nn.Module): 82 | # Focus wh information into c-space 83 | def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True): # ch_in, ch_out, kernel, stride, padding, groups 84 | super(Focus, self).__init__() 85 | self.conv = Conv(c1 * 4, c2, k, s, p, g, act) 86 | 87 | def forward(self, x): # x(b,c,w,h) -> y(b,4c,w/2,h/2) 88 | return self.conv(torch.cat([x[..., ::2, ::2], x[..., 1::2, ::2], x[..., ::2, 1::2], x[..., 1::2, 1::2]], 1)) 89 | 90 | 91 | class Concat(nn.Module): 92 | # Concatenate a list of tensors along dimension 93 | def __init__(self, dimension=1): 94 | super(Concat, self).__init__() 95 | self.d = dimension 96 | 97 | def forward(self, x): 98 | return torch.cat(x, self.d) 99 | 100 | 101 | class Flatten(nn.Module): 102 | # Use after nn.AdaptiveAvgPool2d(1) to remove last 2 dimensions 103 | @staticmethod 104 | def forward(x): 105 | return x.view(x.size(0), -1) 106 | 107 | 108 | class Classify(nn.Module): 109 | # Classification head, i.e. x(b,c1,20,20) to x(b,c2) 110 | def __init__(self, c1, c2, k=1, s=1, p=None, g=1): # ch_in, ch_out, kernel, stride, padding, groups 111 | super(Classify, self).__init__() 112 | self.aap = nn.AdaptiveAvgPool2d(1) # to x(b,c1,1,1) 113 | self.conv = nn.Conv2d(c1, c2, k, s, autopad(k, p), groups=g, bias=False) # to x(b,c2,1,1) 114 | self.flat = Flatten() 115 | 116 | def forward(self, x): 117 | z = torch.cat([self.aap(y) for y in (x if isinstance(x, list) else [x])], 1) # cat if list 118 | return self.flat(self.conv(z)) # flatten to x(b,c2) 119 | -------------------------------------------------------------------------------- /yolov5/models/experimental.py: -------------------------------------------------------------------------------- 1 | # This file contains experimental modules 2 | 3 | import numpy as np 4 | import torch 5 | import torch.nn as nn 6 | 7 | from models.common import Conv, DWConv 8 | from utils.google_utils import attempt_download 9 | 10 | 11 | class CrossConv(nn.Module): 12 | # Cross Convolution Downsample 13 | def __init__(self, c1, c2, k=3, s=1, g=1, e=1.0, shortcut=False): 14 | # ch_in, ch_out, kernel, stride, groups, expansion, shortcut 15 | super(CrossConv, self).__init__() 16 | c_ = int(c2 * e) # hidden channels 17 | self.cv1 = Conv(c1, c_, (1, k), (1, s)) 18 | self.cv2 = Conv(c_, c2, (k, 1), (s, 1), g=g) 19 | self.add = shortcut and c1 == c2 20 | 21 | def forward(self, x): 22 | return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x)) 23 | 24 | 25 | class C3(nn.Module): 26 | # Cross Convolution CSP 27 | def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5): # ch_in, ch_out, number, shortcut, groups, expansion 28 | super(C3, self).__init__() 29 | c_ = int(c2 * e) # hidden channels 30 | self.cv1 = Conv(c1, c_, 1, 1) 31 | self.cv2 = nn.Conv2d(c1, c_, 1, 1, bias=False) 32 | self.cv3 = nn.Conv2d(c_, c_, 1, 1, bias=False) 33 | self.cv4 = Conv(2 * c_, c2, 1, 1) 34 | self.bn = nn.BatchNorm2d(2 * c_) # applied to cat(cv2, cv3) 35 | self.act = nn.LeakyReLU(0.1, inplace=True) 36 | self.m = nn.Sequential(*[CrossConv(c_, c_, 3, 1, g, 1.0, shortcut) for _ in range(n)]) 37 | 38 | def forward(self, x): 39 | y1 = self.cv3(self.m(self.cv1(x))) 40 | y2 = self.cv2(x) 41 | return self.cv4(self.act(self.bn(torch.cat((y1, y2), dim=1)))) 42 | 43 | 44 | class Sum(nn.Module): 45 | # Weighted sum of 2 or more layers https://arxiv.org/abs/1911.09070 46 | def __init__(self, n, weight=False): # n: number of inputs 47 | super(Sum, self).__init__() 48 | self.weight = weight # apply weights boolean 49 | self.iter = range(n - 1) # iter object 50 | if weight: 51 | self.w = nn.Parameter(-torch.arange(1., n) / 2, requires_grad=True) # layer weights 52 | 53 | def forward(self, x): 54 | y = x[0] # no weight 55 | if self.weight: 56 | w = torch.sigmoid(self.w) * 2 57 | for i in self.iter: 58 | y = y + x[i + 1] * w[i] 59 | else: 60 | for i in self.iter: 61 | y = y + x[i + 1] 62 | return y 63 | 64 | 65 | class GhostConv(nn.Module): 66 | # Ghost Convolution https://github.com/huawei-noah/ghostnet 67 | def __init__(self, c1, c2, k=1, s=1, g=1, act=True): # ch_in, ch_out, kernel, stride, groups 68 | super(GhostConv, self).__init__() 69 | c_ = c2 // 2 # hidden channels 70 | self.cv1 = Conv(c1, c_, k, s, g, act) 71 | self.cv2 = Conv(c_, c_, 5, 1, c_, act) 72 | 73 | def forward(self, x): 74 | y = self.cv1(x) 75 | return torch.cat([y, self.cv2(y)], 1) 76 | 77 | 78 | class GhostBottleneck(nn.Module): 79 | # Ghost Bottleneck https://github.com/huawei-noah/ghostnet 80 | def __init__(self, c1, c2, k, s): 81 | super(GhostBottleneck, self).__init__() 82 | c_ = c2 // 2 83 | self.conv = nn.Sequential(GhostConv(c1, c_, 1, 1), # pw 84 | DWConv(c_, c_, k, s, act=False) if s == 2 else nn.Identity(), # dw 85 | GhostConv(c_, c2, 1, 1, act=False)) # pw-linear 86 | self.shortcut = nn.Sequential(DWConv(c1, c1, k, s, act=False), 87 | Conv(c1, c2, 1, 1, act=False)) if s == 2 else nn.Identity() 88 | 89 | def forward(self, x): 90 | return self.conv(x) + self.shortcut(x) 91 | 92 | 93 | class MixConv2d(nn.Module): 94 | # Mixed Depthwise Conv https://arxiv.org/abs/1907.09595 95 | def __init__(self, c1, c2, k=(1, 3), s=1, equal_ch=True): 96 | super(MixConv2d, self).__init__() 97 | groups = len(k) 98 | if equal_ch: # equal c_ per group 99 | i = torch.linspace(0, groups - 1E-6, c2).floor() # c2 indices 100 | c_ = [(i == g).sum() for g in range(groups)] # intermediate channels 101 | else: # equal weight.numel() per group 102 | b = [c2] + [0] * groups 103 | a = np.eye(groups + 1, groups, k=-1) 104 | a -= np.roll(a, 1, axis=1) 105 | a *= np.array(k) ** 2 106 | a[0] = 1 107 | c_ = np.linalg.lstsq(a, b, rcond=None)[0].round() # solve for equal weight indices, ax = b 108 | 109 | self.m = nn.ModuleList([nn.Conv2d(c1, int(c_[g]), k[g], s, k[g] // 2, bias=False) for g in range(groups)]) 110 | self.bn = nn.BatchNorm2d(c2) 111 | self.act = nn.LeakyReLU(0.1, inplace=True) 112 | 113 | def forward(self, x): 114 | return x + self.act(self.bn(torch.cat([m(x) for m in self.m], 1))) 115 | 116 | 117 | class Ensemble(nn.ModuleList): 118 | # Ensemble of models 119 | def __init__(self): 120 | super(Ensemble, self).__init__() 121 | 122 | def forward(self, x, augment=False): 123 | y = [] 124 | for module in self: 125 | y.append(module(x, augment)[0]) 126 | # y = torch.stack(y).max(0)[0] # max ensemble 127 | # y = torch.cat(y, 1) # nms ensemble 128 | y = torch.stack(y).mean(0) # mean ensemble 129 | return y, None # inference, train output 130 | 131 | 132 | def attempt_load(weights, map_location=None): 133 | # Loads an ensemble of models weights=[a,b,c] or a single model weights=[a] or weights=a 134 | model = Ensemble() 135 | for w in weights if isinstance(weights, list) else [weights]: 136 | attempt_download(w) 137 | model.append(torch.load(w, map_location=map_location)['model'].float().fuse().eval()) # load FP32 model 138 | 139 | if len(model) == 1: 140 | return model[-1] # return model 141 | else: 142 | print('Ensemble created with %s\n' % weights) 143 | for k in ['names', 'stride']: 144 | setattr(model, k, getattr(model[-1], k)) 145 | return model # return ensemble 146 | -------------------------------------------------------------------------------- /yolov5/models/export.py: -------------------------------------------------------------------------------- 1 | """Exports a YOLOv5 *.pt model to ONNX and TorchScript formats 2 | 3 | Usage: 4 | $ export PYTHONPATH="$PWD" && python models/export.py --weights ./weights/yolov5s.pt --img 640 --batch 1 5 | """ 6 | 7 | import argparse 8 | 9 | import torch 10 | import torch.nn as nn 11 | 12 | from models.common import Conv 13 | from models.experimental import attempt_load 14 | from utils.activations import Hardswish 15 | from utils.general import set_logging 16 | 17 | if __name__ == '__main__': 18 | parser = argparse.ArgumentParser() 19 | parser.add_argument('--weights', type=str, default='./yolov5s.pt', help='weights path') # from yolov5/models/ 20 | parser.add_argument('--img-size', nargs='+', type=int, default=[640, 640], help='image size') # height, width 21 | parser.add_argument('--batch-size', type=int, default=1, help='batch size') 22 | opt = parser.parse_args() 23 | opt.img_size *= 2 if len(opt.img_size) == 1 else 1 # expand 24 | print(opt) 25 | set_logging() 26 | 27 | # Input 28 | img = torch.zeros((opt.batch_size, 3, *opt.img_size)) # image size(1,3,320,192) iDetection 29 | 30 | # Load PyTorch model 31 | model = attempt_load(opt.weights, map_location=torch.device('cpu')) # load FP32 model 32 | 33 | # Update model 34 | for k, m in model.named_modules(): 35 | m._non_persistent_buffers_set = set() # pytorch 1.6.0 compatability 36 | if isinstance(m, Conv) and isinstance(m.act, nn.Hardswish): 37 | m.act = Hardswish() # assign activation 38 | # if isinstance(m, Detect): 39 | # m.forward = m.forward_export # assign forward (optional) 40 | model.model[-1].export = True # set Detect() layer export=True 41 | y = model(img) # dry run 42 | 43 | # TorchScript export 44 | try: 45 | print('\nStarting TorchScript export with torch %s...' % torch.__version__) 46 | f = opt.weights.replace('.pt', '.torchscript.pt') # filename 47 | ts = torch.jit.trace(model, img) 48 | ts.save(f) 49 | print('TorchScript export success, saved as %s' % f) 50 | except Exception as e: 51 | print('TorchScript export failure: %s' % e) 52 | 53 | # ONNX export 54 | try: 55 | import onnx 56 | 57 | print('\nStarting ONNX export with onnx %s...' % onnx.__version__) 58 | f = opt.weights.replace('.pt', '.onnx') # filename 59 | torch.onnx.export(model, img, f, verbose=False, opset_version=12, input_names=['images'], 60 | output_names=['classes', 'boxes'] if y is None else ['output']) 61 | 62 | # Checks 63 | onnx_model = onnx.load(f) # load onnx model 64 | onnx.checker.check_model(onnx_model) # check onnx model 65 | # print(onnx.helper.printable_graph(onnx_model.graph)) # print a human readable model 66 | print('ONNX export success, saved as %s' % f) 67 | except Exception as e: 68 | print('ONNX export failure: %s' % e) 69 | 70 | # CoreML export 71 | try: 72 | import coremltools as ct 73 | 74 | print('\nStarting CoreML export with coremltools %s...' % ct.__version__) 75 | # convert model from torchscript and apply pixel scaling as per detect.py 76 | model = ct.convert(ts, inputs=[ct.ImageType(name='images', shape=img.shape, scale=1 / 255.0, bias=[0, 0, 0])]) 77 | f = opt.weights.replace('.pt', '.mlmodel') # filename 78 | model.save(f) 79 | print('CoreML export success, saved as %s' % f) 80 | except Exception as e: 81 | print('CoreML export failure: %s' % e) 82 | 83 | # Finish 84 | print('\nExport complete. Visualize with https://github.com/lutzroeder/netron.') 85 | -------------------------------------------------------------------------------- /yolov5/models/hub/yolov3-spp.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.0 # model depth multiple 4 | width_multiple: 1.0 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # darknet53 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Conv, [32, 3, 1]], # 0 16 | [-1, 1, Conv, [64, 3, 2]], # 1-P1/2 17 | [-1, 1, Bottleneck, [64]], 18 | [-1, 1, Conv, [128, 3, 2]], # 3-P2/4 19 | [-1, 2, Bottleneck, [128]], 20 | [-1, 1, Conv, [256, 3, 2]], # 5-P3/8 21 | [-1, 8, Bottleneck, [256]], 22 | [-1, 1, Conv, [512, 3, 2]], # 7-P4/16 23 | [-1, 8, Bottleneck, [512]], 24 | [-1, 1, Conv, [1024, 3, 2]], # 9-P5/32 25 | [-1, 4, Bottleneck, [1024]], # 10 26 | ] 27 | 28 | # YOLOv3-SPP head 29 | head: 30 | [[-1, 1, Bottleneck, [1024, False]], 31 | [-1, 1, SPP, [512, [5, 9, 13]]], 32 | [-1, 1, Conv, [1024, 3, 1]], 33 | [-1, 1, Conv, [512, 1, 1]], 34 | [-1, 1, Conv, [1024, 3, 1]], # 15 (P5/32-large) 35 | 36 | [-2, 1, Conv, [256, 1, 1]], 37 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 38 | [[-1, 8], 1, Concat, [1]], # cat backbone P4 39 | [-1, 1, Bottleneck, [512, False]], 40 | [-1, 1, Bottleneck, [512, False]], 41 | [-1, 1, Conv, [256, 1, 1]], 42 | [-1, 1, Conv, [512, 3, 1]], # 22 (P4/16-medium) 43 | 44 | [-2, 1, Conv, [128, 1, 1]], 45 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 46 | [[-1, 6], 1, Concat, [1]], # cat backbone P3 47 | [-1, 1, Bottleneck, [256, False]], 48 | [-1, 2, Bottleneck, [256, False]], # 27 (P3/8-small) 49 | 50 | [[27, 22, 15], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 51 | ] 52 | -------------------------------------------------------------------------------- /yolov5/models/hub/yolov5-fpn.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.0 # model depth multiple 4 | width_multiple: 1.0 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, Bottleneck, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, BottleneckCSP, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, BottleneckCSP, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 6, BottleneckCSP, [1024]], # 9 25 | ] 26 | 27 | # YOLOv5 FPN head 28 | head: 29 | [[-1, 3, BottleneckCSP, [1024, False]], # 10 (P5/32-large) 30 | 31 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 32 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 33 | [-1, 1, Conv, [512, 1, 1]], 34 | [-1, 3, BottleneckCSP, [512, False]], # 14 (P4/16-medium) 35 | 36 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 37 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 38 | [-1, 1, Conv, [256, 1, 1]], 39 | [-1, 3, BottleneckCSP, [256, False]], # 18 (P3/8-small) 40 | 41 | [[18, 14, 10], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 42 | ] 43 | -------------------------------------------------------------------------------- /yolov5/models/hub/yolov5-panet.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.0 # model depth multiple 4 | width_multiple: 1.0 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [116,90, 156,198, 373,326] # P5/32 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [10,13, 16,30, 33,23] # P3/8 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, BottleneckCSP, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, BottleneckCSP, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, BottleneckCSP, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, BottleneckCSP, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 PANet head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, BottleneckCSP, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, BottleneckCSP, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, BottleneckCSP, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, BottleneckCSP, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P5, P4, P3) 48 | ] 49 | -------------------------------------------------------------------------------- /yolov5/models/yolov5l.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.0 # model depth multiple 4 | width_multiple: 1.0 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, BottleneckCSP, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, BottleneckCSP, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, BottleneckCSP, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, BottleneckCSP, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, BottleneckCSP, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, BottleneckCSP, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, BottleneckCSP, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, BottleneckCSP, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /yolov5/models/yolov5m.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 0.67 # model depth multiple 4 | width_multiple: 0.75 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, BottleneckCSP, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, BottleneckCSP, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, BottleneckCSP, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, BottleneckCSP, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, BottleneckCSP, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, BottleneckCSP, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, BottleneckCSP, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, BottleneckCSP, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /yolov5/models/yolov5s.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 0.33 # model depth multiple 4 | width_multiple: 0.50 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, BottleneckCSP, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, BottleneckCSP, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, BottleneckCSP, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, BottleneckCSP, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, BottleneckCSP, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, BottleneckCSP, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, BottleneckCSP, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, BottleneckCSP, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /yolov5/models/yolov5x.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.33 # model depth multiple 4 | width_multiple: 1.25 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, BottleneckCSP, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, BottleneckCSP, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, BottleneckCSP, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, BottleneckCSP, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, BottleneckCSP, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, BottleneckCSP, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, BottleneckCSP, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, BottleneckCSP, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /yolov5/requirements.txt: -------------------------------------------------------------------------------- 1 | # pip install -r requirements.txt 2 | 3 | # base ---------------------------------------- 4 | Cython 5 | matplotlib>=3.2.2 6 | numpy>=1.18.5 7 | opencv-python>=4.1.2 8 | pillow 9 | PyYAML>=5.3 10 | scipy>=1.4.1 11 | tensorboard>=2.2 12 | torch>=1.6.0 13 | torchvision>=0.7.0 14 | tqdm>=4.41.0 15 | 16 | # coco ---------------------------------------- 17 | # pycocotools>=2.0 18 | 19 | # export -------------------------------------- 20 | # packaging # for coremltools 21 | # coremltools==4.0b3 22 | # onnx>=1.7.0 23 | # scikit-learn==0.19.2 # for coreml quantization 24 | 25 | # extras -------------------------------------- 26 | # thop 27 | -------------------------------------------------------------------------------- /yolov5/utils/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/BioMeasure/PyQt5_YoLoV5_DeepSort/4510632b9ed8dcb22c19f4ef87588f1b42d2575c/yolov5/utils/__init__.py -------------------------------------------------------------------------------- /yolov5/utils/activations.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn as nn 3 | import torch.nn.functional as F 4 | 5 | 6 | # Swish https://arxiv.org/pdf/1905.02244.pdf --------------------------------------------------------------------------- 7 | class Swish(nn.Module): # 8 | @staticmethod 9 | def forward(x): 10 | return x * torch.sigmoid(x) 11 | 12 | 13 | class Hardswish(nn.Module): # export-friendly version of nn.Hardswish() 14 | @staticmethod 15 | def forward(x): 16 | # return x * F.hardsigmoid(x) # for torchscript and CoreML 17 | return x * F.hardtanh(x + 3, 0., 6.) / 6. # for torchscript, CoreML and ONNX 18 | 19 | 20 | class MemoryEfficientSwish(nn.Module): 21 | class F(torch.autograd.Function): 22 | @staticmethod 23 | def forward(ctx, x): 24 | ctx.save_for_backward(x) 25 | return x * torch.sigmoid(x) 26 | 27 | @staticmethod 28 | def backward(ctx, grad_output): 29 | x = ctx.saved_tensors[0] 30 | sx = torch.sigmoid(x) 31 | return grad_output * (sx * (1 + x * (1 - sx))) 32 | 33 | def forward(self, x): 34 | return self.F.apply(x) 35 | 36 | 37 | # Mish https://github.com/digantamisra98/Mish -------------------------------------------------------------------------- 38 | class Mish(nn.Module): 39 | @staticmethod 40 | def forward(x): 41 | return x * F.softplus(x).tanh() 42 | 43 | 44 | class MemoryEfficientMish(nn.Module): 45 | class F(torch.autograd.Function): 46 | @staticmethod 47 | def forward(ctx, x): 48 | ctx.save_for_backward(x) 49 | return x.mul(torch.tanh(F.softplus(x))) # x * tanh(ln(1 + exp(x))) 50 | 51 | @staticmethod 52 | def backward(ctx, grad_output): 53 | x = ctx.saved_tensors[0] 54 | sx = torch.sigmoid(x) 55 | fx = F.softplus(x).tanh() 56 | return grad_output * (fx + x * sx * (1 - fx * fx)) 57 | 58 | def forward(self, x): 59 | return self.F.apply(x) 60 | 61 | 62 | # FReLU https://arxiv.org/abs/2007.11824 ------------------------------------------------------------------------------- 63 | class FReLU(nn.Module): 64 | def __init__(self, c1, k=3): # ch_in, kernel 65 | super().__init__() 66 | self.conv = nn.Conv2d(c1, c1, k, 1, 1, groups=c1) 67 | self.bn = nn.BatchNorm2d(c1) 68 | 69 | def forward(self, x): 70 | return torch.max(x, self.bn(self.conv(x))) 71 | -------------------------------------------------------------------------------- /yolov5/utils/evolve.sh: -------------------------------------------------------------------------------- 1 | #!/bin/bash 2 | # Hyperparameter evolution commands (avoids CUDA memory leakage issues) 3 | # Replaces train.py python generations 'for' loop with a bash 'for' loop 4 | 5 | # Start on 4-GPU machine 6 | #for i in 0 1 2 3; do 7 | # t=ultralytics/yolov5:test && sudo docker pull $t && sudo docker run -d --ipc=host --gpus all -v "$(pwd)"/VOC:/usr/src/VOC $t bash utils/evolve.sh $i 8 | # sleep 60 # avoid simultaneous evolve.txt read/write 9 | #done 10 | 11 | # Hyperparameter evolution commands 12 | while true; do 13 | python train.py --batch 64 --weights yolov5m.pt --data voc.yaml --img 512 --epochs 50 --evolve --bucket ult/voc --device $1 14 | done 15 | -------------------------------------------------------------------------------- /yolov5/utils/google_utils.py: -------------------------------------------------------------------------------- 1 | # This file contains google utils: https://cloud.google.com/storage/docs/reference/libraries 2 | # pip install --upgrade google-cloud-storage 3 | # from google.cloud import storage 4 | 5 | import os 6 | import platform 7 | import time 8 | from pathlib import Path 9 | 10 | import torch 11 | 12 | 13 | def attempt_download(weights): 14 | # Attempt to download pretrained weights if not found locally 15 | weights = weights.strip().replace("'", '') 16 | file = Path(weights).name 17 | 18 | msg = weights + ' missing, try downloading from https://github.com/ultralytics/yolov5/releases/' 19 | models = ['yolov5s.pt', 'yolov5m.pt', 'yolov5l.pt', 'yolov5x.pt'] # available models 20 | 21 | if file in models and not os.path.isfile(weights): 22 | # Google Drive 23 | # d = {'yolov5s.pt': '1R5T6rIyy3lLwgFXNms8whc-387H0tMQO', 24 | # 'yolov5m.pt': '1vobuEExpWQVpXExsJ2w-Mbf3HJjWkQJr', 25 | # 'yolov5l.pt': '1hrlqD1Wdei7UT4OgT785BEk1JwnSvNEV', 26 | # 'yolov5x.pt': '1mM8aZJlWTxOg7BZJvNUMrTnA2AbeCVzS'} 27 | # r = gdrive_download(id=d[file], name=weights) if file in d else 1 28 | # if r == 0 and os.path.exists(weights) and os.path.getsize(weights) > 1E6: # check 29 | # return 30 | 31 | try: # GitHub 32 | url = 'https://github.com/ultralytics/yolov5/releases/download/v3.0/' + file 33 | print('Downloading %s to %s...' % (url, weights)) 34 | if platform.system() == 'Darwin': # avoid MacOS python requests certificate error 35 | r = os.system('curl -L %s -o %s' % (url, weights)) 36 | else: 37 | torch.hub.download_url_to_file(url, weights) 38 | assert os.path.exists(weights) and os.path.getsize(weights) > 1E6 # check 39 | except Exception as e: # GCP 40 | print('Download error: %s' % e) 41 | url = 'https://storage.googleapis.com/ultralytics/yolov5/ckpt/' + file 42 | print('Downloading %s to %s...' % (url, weights)) 43 | r = os.system('curl -L %s -o %s' % (url, weights)) # torch.hub.download_url_to_file(url, weights) 44 | finally: 45 | if not (os.path.exists(weights) and os.path.getsize(weights) > 1E6): # check 46 | os.remove(weights) if os.path.exists(weights) else None # remove partial downloads 47 | print('ERROR: Download failure: %s' % msg) 48 | print('') 49 | return 50 | 51 | 52 | def gdrive_download(id='1n_oKgR81BJtqk75b00eAjdv03qVCQn2f', name='coco128.zip'): 53 | # Downloads a file from Google Drive. from utils.google_utils import *; gdrive_download() 54 | t = time.time() 55 | 56 | print('Downloading https://drive.google.com/uc?export=download&id=%s as %s... ' % (id, name), end='') 57 | os.remove(name) if os.path.exists(name) else None # remove existing 58 | os.remove('cookie') if os.path.exists('cookie') else None 59 | 60 | # Attempt file download 61 | out = "NUL" if platform.system() == "Windows" else "/dev/null" 62 | os.system('curl -c ./cookie -s -L "drive.google.com/uc?export=download&id=%s" > %s ' % (id, out)) 63 | if os.path.exists('cookie'): # large file 64 | s = 'curl -Lb ./cookie "drive.google.com/uc?export=download&confirm=%s&id=%s" -o %s' % (get_token(), id, name) 65 | else: # small file 66 | s = 'curl -s -L -o %s "drive.google.com/uc?export=download&id=%s"' % (name, id) 67 | r = os.system(s) # execute, capture return 68 | os.remove('cookie') if os.path.exists('cookie') else None 69 | 70 | # Error check 71 | if r != 0: 72 | os.remove(name) if os.path.exists(name) else None # remove partial 73 | print('Download error ') # raise Exception('Download error') 74 | return r 75 | 76 | # Unzip if archive 77 | if name.endswith('.zip'): 78 | print('unzipping... ', end='') 79 | os.system('unzip -q %s' % name) # unzip 80 | os.remove(name) # remove zip to free space 81 | 82 | print('Done (%.1fs)' % (time.time() - t)) 83 | return r 84 | 85 | 86 | def get_token(cookie="./cookie"): 87 | with open(cookie) as f: 88 | for line in f: 89 | if "download" in line: 90 | return line.split()[-1] 91 | return "" 92 | 93 | # def upload_blob(bucket_name, source_file_name, destination_blob_name): 94 | # # Uploads a file to a bucket 95 | # # https://cloud.google.com/storage/docs/uploading-objects#storage-upload-object-python 96 | # 97 | # storage_client = storage.Client() 98 | # bucket = storage_client.get_bucket(bucket_name) 99 | # blob = bucket.blob(destination_blob_name) 100 | # 101 | # blob.upload_from_filename(source_file_name) 102 | # 103 | # print('File {} uploaded to {}.'.format( 104 | # source_file_name, 105 | # destination_blob_name)) 106 | # 107 | # 108 | # def download_blob(bucket_name, source_blob_name, destination_file_name): 109 | # # Uploads a blob from a bucket 110 | # storage_client = storage.Client() 111 | # bucket = storage_client.get_bucket(bucket_name) 112 | # blob = bucket.blob(source_blob_name) 113 | # 114 | # blob.download_to_filename(destination_file_name) 115 | # 116 | # print('Blob {} downloaded to {}.'.format( 117 | # source_blob_name, 118 | # destination_file_name)) 119 | -------------------------------------------------------------------------------- /yolov5/weights/download_weights.sh: -------------------------------------------------------------------------------- 1 | #!/bin/bash 2 | # Download common models 3 | 4 | python -c " 5 | from utils.google_utils import *; 6 | attempt_download('weights/yolov5s.pt'); 7 | attempt_download('weights/yolov5m.pt'); 8 | attempt_download('weights/yolov5l.pt'); 9 | attempt_download('weights/yolov5x.pt') 10 | " 11 | --------------------------------------------------------------------------------