├── .gitignore ├── .idea ├── .gitignore ├── inspectionProfiles │ └── profiles_settings.xml ├── misc.xml ├── modules.xml ├── other.xml ├── vcs.xml └── win10_yolov5_deepsort_counting.iml ├── Arial.ttf ├── README.md ├── cover.png ├── deep_sort ├── configs │ └── deep_sort.yaml ├── deep_sort │ ├── README.md │ ├── __init__.py │ ├── deep │ │ ├── __init__.py │ │ ├── checkpoint │ │ │ ├── .gitkeep │ │ │ └── ckpt.t7 │ │ ├── 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 └── utils │ ├── __init__.py │ ├── asserts.py │ ├── draw.py │ ├── evaluation.py │ ├── io.py │ ├── json_logger.py │ ├── log.py │ ├── parser.py │ └── tools.py ├── detector.py ├── main.py ├── models ├── __init__.py ├── common.py ├── experimental.py ├── hub │ ├── anchors.yaml │ ├── yolov3-spp.yaml │ ├── yolov3-tiny.yaml │ ├── yolov3.yaml │ ├── yolov5-bifpn.yaml │ ├── yolov5-fpn.yaml │ ├── yolov5-p2.yaml │ ├── yolov5-p34.yaml │ ├── yolov5-p6.yaml │ ├── yolov5-p7.yaml │ ├── yolov5-panet.yaml │ ├── yolov5l6.yaml │ ├── yolov5m6.yaml │ ├── yolov5n6.yaml │ ├── yolov5s-ghost.yaml │ ├── yolov5s-transformer.yaml │ ├── yolov5s6.yaml │ └── yolov5x6.yaml ├── tf.py ├── yolo.py ├── yolov5l.yaml ├── yolov5m.yaml ├── yolov5n.yaml ├── yolov5s.yaml └── yolov5x.yaml ├── requirements.txt ├── tracker.py ├── utils ├── __init__.py ├── activations.py ├── augmentations.py ├── autoanchor.py ├── autobatch.py ├── aws │ ├── __init__.py │ ├── mime.sh │ ├── resume.py │ └── userdata.sh ├── benchmarks.py ├── callbacks.py ├── datasets.py ├── downloads.py ├── flask_rest_api │ ├── README.md │ ├── example_request.py │ └── restapi.py ├── general.py ├── google_app_engine │ ├── Dockerfile │ ├── additional_requirements.txt │ └── app.yaml ├── loggers │ ├── __init__.py │ └── wandb │ │ ├── README.md │ │ ├── __init__.py │ │ ├── log_dataset.py │ │ ├── sweep.py │ │ ├── sweep.yaml │ │ └── wandb_utils.py ├── loss.py ├── metrics.py ├── plots.py └── torch_utils.py ├── video └── test.mp4 └── weights └── yolov5m.pt /.gitignore: -------------------------------------------------------------------------------- 1 | # Byte-compiled / optimized / DLL files 2 | __pycache__/ 3 | *.py[cod] 4 | *$py.class 5 | 6 | # C extensions 7 | *.so 8 | 9 | # Distribution / packaging 10 | .Python 11 | build/ 12 | develop-eggs/ 13 | dist/ 14 | downloads/ 15 | eggs/ 16 | .eggs/ 17 | lib/ 18 | lib64/ 19 | parts/ 20 | sdist/ 21 | var/ 22 | wheels/ 23 | pip-wheel-metadata/ 24 | share/python-wheels/ 25 | *.egg-info/ 26 | .installed.cfg 27 | *.egg 28 | MANIFEST 29 | 30 | # PyInstaller 31 | # Usually these files are written by a python script from a template 32 | # before PyInstaller builds the exe, so as to inject date/other infos into it. 33 | *.manifest 34 | *.spec 35 | 36 | # Installer logs 37 | pip-log.txt 38 | pip-delete-this-directory.txt 39 | 40 | # Unit test / coverage reports 41 | htmlcov/ 42 | .tox/ 43 | .nox/ 44 | .coverage 45 | .coverage.* 46 | .cache 47 | nosetests.xml 48 | coverage.xml 49 | *.cover 50 | *.py,cover 51 | .hypothesis/ 52 | .pytest_cache/ 53 | 54 | # Translations 55 | *.mo 56 | *.pot 57 | 58 | # Django stuff: 59 | *.log 60 | local_settings.py 61 | db.sqlite3 62 | db.sqlite3-journal 63 | 64 | # Flask stuff: 65 | instance/ 66 | .webassets-cache 67 | 68 | # Scrapy stuff: 69 | .scrapy 70 | 71 | # Sphinx documentation 72 | docs/_build/ 73 | 74 | # PyBuilder 75 | target/ 76 | 77 | # Jupyter Notebook 78 | .ipynb_checkpoints 79 | 80 | # IPython 81 | profile_default/ 82 | ipython_config.py 83 | 84 | # pyenv 85 | .python-version 86 | 87 | # pipenv 88 | # According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control. 89 | # However, in case of collaboration, if having platform-specific dependencies or dependencies 90 | # having no cross-platform support, pipenv may install dependencies that don't work, or not 91 | # install all needed dependencies. 92 | #Pipfile.lock 93 | 94 | # PEP 582; used by e.g. github.com/David-OConnor/pyflow 95 | __pypackages__/ 96 | 97 | # Celery stuff 98 | celerybeat-schedule 99 | celerybeat.pid 100 | 101 | # SageMath parsed files 102 | *.sage.py 103 | 104 | # Environments 105 | .env 106 | .venv 107 | env/ 108 | venv/ 109 | ENV/ 110 | env.bak/ 111 | venv.bak/ 112 | 113 | # Spyder project settings 114 | .spyderproject 115 | .spyproject 116 | 117 | # Rope project settings 118 | .ropeproject 119 | 120 | # mkdocs documentation 121 | /site 122 | 123 | # mypy 124 | .mypy_cache/ 125 | .dmypy.json 126 | dmypy.json 127 | 128 | # Pyre type checker 129 | .pyre/ 130 | -------------------------------------------------------------------------------- /.idea/.gitignore: -------------------------------------------------------------------------------- 1 | # Default ignored files 2 | /shelf/ 3 | /workspace.xml 4 | # Editor-based HTTP Client requests 5 | /httpRequests/ 6 | # Datasource local storage ignored files 7 | /dataSources/ 8 | /dataSources.local.xml 9 | -------------------------------------------------------------------------------- /.idea/inspectionProfiles/profiles_settings.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 6 | -------------------------------------------------------------------------------- /.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 | -------------------------------------------------------------------------------- /.idea/win10_yolov5_deepsort_counting.iml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 12 | -------------------------------------------------------------------------------- /Arial.ttf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/dyh/win10_yolov5_deepsort_counting/adb686ee839e89177990c27f5da35bfae7ab4b9b/Arial.ttf -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # win10版本 yolov5 deepsort 行人 车辆 跟踪 检测 计数 2 | 3 | ## 应B站上同学们要求在 win10 运行 4 | 5 | - 更新到 python 3.9.10，请不要安装更高版本。 6 | - 更新到 yolov5 v6.1版。使用的权重文件可以在此下载：https://github.com/ultralytics/yolov5/releases/tag/v6.1 7 | - 更新到 CUDA 11.3+ 8 | - 建议保留 Arial.ttf 文件，或在首次运行时由 yolov5 自动下载。 9 | 10 | 11 | ## 功能 12 | - 实现了 出/入 分别计数。 13 | - 显示检测类别。 14 | - 默认是 南/北 方向检测，若要检测不同位置和方向，可在 main.py 文件第13行和21行，修改2个polygon的点。 15 | - 默认检测类别：行人、自行车、小汽车、摩托车、公交车、卡车。 16 | - 检测类别可在 detector.py 文件第60行修改。 17 | 18 | 19 | ### 视频 20 | 21 | bilibili 22 | 23 | [![bilibili](https://raw.githubusercontent.com/dyh/win10_yolov5_deepsort_counting/main/cover.png)](https://www.bilibili.com/video/BV13Z4y1C7Dt/ "bilibili") 24 | 25 | 26 | ## 运行环境 27 | 28 | - python 3.9.10，pip 22.0.3+ 29 | - pytorch 1.10.2+ 30 | - pip3 install -r requirements.txt 31 | 32 | 33 | ## 如何运行 34 | 35 | 0. 确保正确安装 python 和 CUDA 36 | 37 | ``` 38 | D:\> python -V 39 | 40 | D:\> nvidia-smi 41 | 42 | D:\> nvcc -V 43 | ``` 44 | 45 | 1. 下载代码 46 | 47 | ``` 48 | D:\> git clone https://github.com/dyh/win10_yolov5_deepsort_counting.git 49 | ``` 50 | 51 | > 因此repo包含weights和mp4文件，若 git clone 速度慢，可直接下载zip文件：https://github.com/dyh/win10_yolov5_deepsort_counting/archive/refs/heads/main.zip 52 | 53 | 2. 进入目录 54 | 55 | ``` 56 | D:\> cd win10_yolov5_deepsort_counting 57 | ``` 58 | 59 | 3. 创建 python 虚拟环境 60 | 61 | ``` 62 | D:\win10_yolov5_deepsort_counting> python -m venv venv 63 | ``` 64 | 65 | 4. 激活虚拟环境 66 | 67 | ``` 68 | D:\win10_yolov5_deepsort_counting> venv\Scripts\activate 69 | ``` 70 | 71 | 5. 升级pip 72 | 73 | ``` 74 | (venv) D:\win10_yolov5_deepsort_counting> python -m pip install --upgrade pip 75 | ``` 76 | 77 | 6. 安装pytorch 78 | 79 | > 根据你的操作系统、虚拟环境以及CUDA版本，在 https://pytorch.org/get-started/locally/ 找到对应的安装命令。我的环境是 win10、pip、CUDA 11.6。 80 | 81 | ``` 82 | (venv) D:\win10_yolov5_deepsort_counting> pip3 install torch==1.10.2+cu113 torchvision==0.11.3+cu113 torchaudio===0.10.2+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html 83 | ``` 84 | 85 | 7. 安装软件包 86 | 87 | ``` 88 | (venv) D:\win10_yolov5_deepsort_counting> pip3 install -r requirements.txt 89 | ``` 90 | 91 | 8. 在 main.py 文件中第66行，设置要检测的视频文件路径，默认为 './video/test.mp4' 92 | 93 | > 140MB的测试视频可以在这里下载：https://pan.baidu.com/s/1qHNGGpX1QD6zHyNTqWvg1w 提取码: 8ufq 94 | 95 | ``` 96 | capture = cv2.VideoCapture(r'video\test.mp4') 97 | ``` 98 | 99 | 9. 运行程序 100 | 101 | ``` 102 | (venv) D:\win10_yolov5_deepsort_counting> python main.py 103 | ``` 104 | 105 | 106 | ## 使用框架 107 | 108 | - https://github.com/Sharpiless/Yolov5-deepsort-inference 109 | - https://github.com/ultralytics/yolov5/ 110 | - https://github.com/ZQPei/deep_sort_pytorch 111 | -------------------------------------------------------------------------------- /cover.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/dyh/win10_yolov5_deepsort_counting/adb686ee839e89177990c27f5da35bfae7ab4b9b/cover.png -------------------------------------------------------------------------------- /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/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/dyh/win10_yolov5_deepsort_counting/adb686ee839e89177990c27f5da35bfae7ab4b9b/deep_sort/deep_sort/deep/__init__.py -------------------------------------------------------------------------------- /deep_sort/deep_sort/deep/checkpoint/.gitkeep: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/dyh/win10_yolov5_deepsort_counting/adb686ee839e89177990c27f5da35bfae7ab4b9b/deep_sort/deep_sort/deep/checkpoint/.gitkeep -------------------------------------------------------------------------------- /deep_sort/deep_sort/deep/checkpoint/ckpt.t7: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/dyh/win10_yolov5_deepsort_counting/adb686ee839e89177990c27f5da35bfae7ab4b9b/deep_sort/deep_sort/deep/checkpoint/ckpt.t7 -------------------------------------------------------------------------------- /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/dyh/win10_yolov5_deepsort_counting/adb686ee839e89177990c27f5da35bfae7ab4b9b/deep_sort/deep_sort/deep/train.jpg -------------------------------------------------------------------------------- /deep_sort/deep_sort/deep/train.py: -------------------------------------------------------------------------------- 1 | import argparse 2 | import os 3 | import time 4 | 5 | import numpy as np 6 | import matplotlib.pyplot as plt 7 | import torch 8 | import torch.backends.cudnn as cudnn 9 | import torchvision 10 | 11 | from model import Net 12 | 13 | parser = argparse.ArgumentParser(description="Train on market1501") 14 | parser.add_argument("--data-dir",default='data',type=str) 15 | parser.add_argument("--no-cuda",action="store_true") 16 | parser.add_argument("--gpu-id",default=0,type=int) 17 | parser.add_argument("--lr",default=0.1, type=float) 18 | parser.add_argument("--interval",'-i',default=20,type=int) 19 | parser.add_argument('--resume', '-r',action='store_true') 20 | args = parser.parse_args() 21 | 22 | # device 23 | device = "cuda:{}".format(args.gpu_id) if torch.cuda.is_available() and not args.no_cuda else "cpu" 24 | if torch.cuda.is_available() and not args.no_cuda: 25 | cudnn.benchmark = True 26 | 27 | # data loading 28 | root = args.data_dir 29 | train_dir = os.path.join(root,"train") 30 | test_dir = os.path.join(root,"test") 31 | transform_train = torchvision.transforms.Compose([ 32 | torchvision.transforms.RandomCrop((128,64),padding=4), 33 | torchvision.transforms.RandomHorizontalFlip(), 34 | torchvision.transforms.ToTensor(), 35 | torchvision.transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) 36 | ]) 37 | transform_test = torchvision.transforms.Compose([ 38 | torchvision.transforms.Resize((128,64)), 39 | torchvision.transforms.ToTensor(), 40 | torchvision.transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) 41 | ]) 42 | trainloader = torch.utils.data.DataLoader( 43 | torchvision.datasets.ImageFolder(train_dir, transform=transform_train), 44 | batch_size=64,shuffle=True 45 | ) 46 | testloader = torch.utils.data.DataLoader( 47 | torchvision.datasets.ImageFolder(test_dir, transform=transform_test), 48 | batch_size=64,shuffle=True 49 | ) 50 | num_classes = max(len(trainloader.dataset.classes), len(testloader.dataset.classes)) 51 | 52 | # net definition 53 | start_epoch = 0 54 | net = Net(num_classes=num_classes) 55 | if args.resume: 56 | assert os.path.isfile("./checkpoint/ckpt.t7"), "Error: no checkpoint file found!" 57 | print('Loading from checkpoint/ckpt.t7') 58 | checkpoint = torch.load("./checkpoint/ckpt.t7") 59 | # import ipdb; ipdb.set_trace() 60 | net_dict = checkpoint['net_dict'] 61 | net.load_state_dict(net_dict) 62 | best_acc = checkpoint['acc'] 63 | start_epoch = checkpoint['epoch'] 64 | net.to(device) 65 | 66 | # loss and optimizer 67 | criterion = torch.nn.CrossEntropyLoss() 68 | optimizer = torch.optim.SGD(net.parameters(), args.lr, momentum=0.9, weight_decay=5e-4) 69 | best_acc = 0. 70 | 71 | # train function for each epoch 72 | def train(epoch): 73 | print("\nEpoch : %d"%(epoch+1)) 74 | net.train() 75 | training_loss = 0. 76 | train_loss = 0. 77 | correct = 0 78 | total = 0 79 | interval = args.interval 80 | start = time.time() 81 | for idx, (inputs, labels) in enumerate(trainloader): 82 | # forward 83 | inputs,labels = inputs.to(device),labels.to(device) 84 | outputs = net(inputs) 85 | loss = criterion(outputs, labels) 86 | 87 | # backward 88 | optimizer.zero_grad() 89 | loss.backward() 90 | optimizer.step() 91 | 92 | # accumurating 93 | training_loss += loss.item() 94 | train_loss += loss.item() 95 | correct += outputs.max(dim=1)[1].eq(labels).sum().item() 96 | total += labels.size(0) 97 | 98 | # print 99 | if (idx+1)%interval == 0: 100 | end = time.time() 101 | print("[progress:{:.1f}%]time:{:.2f}s Loss:{:.5f} Correct:{}/{} Acc:{:.3f}%".format( 102 | 100.*(idx+1)/len(trainloader), end-start, training_loss/interval, correct, total, 100.*correct/total 103 | )) 104 | training_loss = 0. 105 | start = time.time() 106 | 107 | return train_loss/len(trainloader), 1.- correct/total 108 | 109 | def test(epoch): 110 | global best_acc 111 | net.eval() 112 | test_loss = 0. 113 | correct = 0 114 | total = 0 115 | start = time.time() 116 | with torch.no_grad(): 117 | for idx, (inputs, labels) in enumerate(testloader): 118 | inputs, labels = inputs.to(device), labels.to(device) 119 | outputs = net(inputs) 120 | loss = criterion(outputs, labels) 121 | 122 | test_loss += loss.item() 123 | correct += outputs.max(dim=1)[1].eq(labels).sum().item() 124 | total += labels.size(0) 125 | 126 | print("Testing ...") 127 | end = time.time() 128 | print("[progress:{:.1f}%]time:{:.2f}s Loss:{:.5f} Correct:{}/{} Acc:{:.3f}%".format( 129 | 100.*(idx+1)/len(testloader), end-start, test_loss/len(testloader), correct, total, 100.*correct/total 130 | )) 131 | 132 | # saving checkpoint 133 | acc = 100.*correct/total 134 | if acc > best_acc: 135 | best_acc = acc 136 | print("Saving parameters to checkpoint/ckpt.t7") 137 | checkpoint = { 138 | 'net_dict':net.state_dict(), 139 | 'acc':acc, 140 | 'epoch':epoch, 141 | } 142 | if not os.path.isdir('checkpoint'): 143 | os.mkdir('checkpoint') 144 | torch.save(checkpoint, './checkpoint/ckpt.t7') 145 | 146 | return test_loss/len(testloader), 1.- correct/total 147 | 148 | # plot figure 149 | x_epoch = [] 150 | record = {'train_loss':[], 'train_err':[], 'test_loss':[], 'test_err':[]} 151 | fig = plt.figure() 152 | ax0 = fig.add_subplot(121, title="loss") 153 | ax1 = fig.add_subplot(122, title="top1err") 154 | def draw_curve(epoch, train_loss, train_err, test_loss, test_err): 155 | global record 156 | record['train_loss'].append(train_loss) 157 | record['train_err'].append(train_err) 158 | record['test_loss'].append(test_loss) 159 | record['test_err'].append(test_err) 160 | 161 | x_epoch.append(epoch) 162 | ax0.plot(x_epoch, record['train_loss'], 'bo-', label='train') 163 | ax0.plot(x_epoch, record['test_loss'], 'ro-', label='val') 164 | ax1.plot(x_epoch, record['train_err'], 'bo-', label='train') 165 | ax1.plot(x_epoch, record['test_err'], 'ro-', label='val') 166 | if epoch == 0: 167 | ax0.legend() 168 | ax1.legend() 169 | fig.savefig("train.jpg") 170 | 171 | # lr decay 172 | def lr_decay(): 173 | global optimizer 174 | for params in optimizer.param_groups: 175 | params['lr'] *= 0.1 176 | lr = params['lr'] 177 | print("Learning rate adjusted to {}".format(lr)) 178 | 179 | def main(): 180 | for epoch in range(start_epoch, start_epoch+40): 181 | train_loss, train_err = train(epoch) 182 | test_loss, test_err = test(epoch) 183 | draw_curve(epoch, train_loss, train_err, test_loss, test_err) 184 | if (epoch+1)%20==0: 185 | lr_decay() 186 | 187 | 188 | if __name__ == '__main__': 189 | main() 190 | -------------------------------------------------------------------------------- /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/dyh/win10_yolov5_deepsort_counting/adb686ee839e89177990c27f5da35bfae7ab4b9b/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/kalman_filter.py: -------------------------------------------------------------------------------- 1 | # vim: expandtab:ts=4:sw=4 2 | import numpy as np 3 | import scipy.linalg 4 | 5 | 6 | """ 7 | Table for the 0.95 quantile of the chi-square distribution with N degrees of 8 | freedom (contains values for N=1, ..., 9). Taken from MATLAB/Octave's chi2inv 9 | function and used as Mahalanobis gating threshold. 10 | """ 11 | chi2inv95 = { 12 | 1: 3.8415, 13 | 2: 5.9915, 14 | 3: 7.8147, 15 | 4: 9.4877, 16 | 5: 11.070, 17 | 6: 12.592, 18 | 7: 14.067, 19 | 8: 15.507, 20 | 9: 16.919} 21 | 22 | 23 | class KalmanFilter(object): 24 | """ 25 | A simple Kalman filter for tracking bounding boxes in image space. 26 | 27 | The 8-dimensional state space 28 | 29 | x, y, a, h, vx, vy, va, vh 30 | 31 | contains the bounding box center position (x, y), aspect ratio a, height h, 32 | and their respective velocities. 33 | 34 | Object motion follows a constant velocity model. The bounding box location 35 | (x, y, a, h) is taken as direct observation of the state space (linear 36 | observation model). 37 | 38 | """ 39 | 40 | def __init__(self): 41 | ndim, dt = 4, 1. 42 | 43 | # Create Kalman filter model matrices. 44 | self._motion_mat = np.eye(2 * ndim, 2 * ndim) 45 | for i in range(ndim): 46 | self._motion_mat[i, ndim + i] = dt 47 | self._update_mat = np.eye(ndim, 2 * ndim) 48 | 49 | # Motion and observation uncertainty are chosen relative to the current 50 | # state estimate. These weights control the amount of uncertainty in 51 | # the model. This is a bit hacky. 52 | self._std_weight_position = 1. / 20 53 | self._std_weight_velocity = 1. / 160 54 | 55 | def initiate(self, measurement): 56 | """Create track from unassociated measurement. 57 | 58 | Parameters 59 | ---------- 60 | measurement : ndarray 61 | Bounding box coordinates (x, y, a, h) with center position (x, y), 62 | aspect ratio a, and height h. 63 | 64 | Returns 65 | ------- 66 | (ndarray, ndarray) 67 | Returns the mean vector (8 dimensional) and covariance matrix (8x8 68 | dimensional) of the new track. Unobserved velocities are initialized 69 | to 0 mean. 70 | 71 | """ 72 | mean_pos = measurement 73 | mean_vel = np.zeros_like(mean_pos) 74 | mean = np.r_[mean_pos, mean_vel] 75 | 76 | std = [ 77 | 2 * self._std_weight_position * measurement[3], 78 | 2 * self._std_weight_position * measurement[3], 79 | 1e-2, 80 | 2 * self._std_weight_position * measurement[3], 81 | 10 * self._std_weight_velocity * measurement[3], 82 | 10 * self._std_weight_velocity * measurement[3], 83 | 1e-5, 84 | 10 * self._std_weight_velocity * measurement[3]] 85 | covariance = np.diag(np.square(std)) 86 | return mean, covariance 87 | 88 | def predict(self, mean, covariance): 89 | """Run Kalman filter prediction step. 90 | 91 | Parameters 92 | ---------- 93 | mean : ndarray 94 | The 8 dimensional mean vector of the object state at the previous 95 | time step. 96 | covariance : ndarray 97 | The 8x8 dimensional covariance matrix of the object state at the 98 | previous time step. 99 | 100 | Returns 101 | ------- 102 | (ndarray, ndarray) 103 | Returns the mean vector and covariance matrix of the predicted 104 | state. Unobserved velocities are initialized to 0 mean. 105 | 106 | """ 107 | std_pos = [ 108 | self._std_weight_position * mean[3], 109 | self._std_weight_position * mean[3], 110 | 1e-2, 111 | self._std_weight_position * mean[3]] 112 | std_vel = [ 113 | self._std_weight_velocity * mean[3], 114 | self._std_weight_velocity * mean[3], 115 | 1e-5, 116 | self._std_weight_velocity * mean[3]] 117 | motion_cov = np.diag(np.square(np.r_[std_pos, std_vel])) 118 | 119 | mean = np.dot(self._motion_mat, mean) 120 | covariance = np.linalg.multi_dot(( 121 | self._motion_mat, covariance, self._motion_mat.T)) + motion_cov 122 | 123 | return mean, covariance 124 | 125 | def project(self, mean, covariance): 126 | """Project state distribution to measurement space. 127 | 128 | Parameters 129 | ---------- 130 | mean : ndarray 131 | The state's mean vector (8 dimensional array). 132 | covariance : ndarray 133 | The state's covariance matrix (8x8 dimensional). 134 | 135 | Returns 136 | ------- 137 | (ndarray, ndarray) 138 | Returns the projected mean and covariance matrix of the given state 139 | estimate. 140 | 141 | """ 142 | std = [ 143 | self._std_weight_position * mean[3], 144 | self._std_weight_position * mean[3], 145 | 1e-1, 146 | self._std_weight_position * mean[3]] 147 | innovation_cov = np.diag(np.square(std)) 148 | 149 | mean = np.dot(self._update_mat, mean) 150 | covariance = np.linalg.multi_dot(( 151 | self._update_mat, covariance, self._update_mat.T)) 152 | return mean, covariance + innovation_cov 153 | 154 | def update(self, mean, covariance, measurement): 155 | """Run Kalman filter correction step. 156 | 157 | Parameters 158 | ---------- 159 | mean : ndarray 160 | The predicted state's mean vector (8 dimensional). 161 | covariance : ndarray 162 | The state's covariance matrix (8x8 dimensional). 163 | measurement : ndarray 164 | The 4 dimensional measurement vector (x, y, a, h), where (x, y) 165 | is the center position, a the aspect ratio, and h the height of the 166 | bounding box. 167 | 168 | Returns 169 | ------- 170 | (ndarray, ndarray) 171 | Returns the measurement-corrected state distribution. 172 | 173 | """ 174 | projected_mean, projected_cov = self.project(mean, covariance) 175 | 176 | chol_factor, lower = scipy.linalg.cho_factor( 177 | projected_cov, lower=True, check_finite=False) 178 | kalman_gain = scipy.linalg.cho_solve( 179 | (chol_factor, lower), np.dot(covariance, self._update_mat.T).T, 180 | check_finite=False).T 181 | innovation = measurement - projected_mean 182 | 183 | new_mean = mean + np.dot(innovation, kalman_gain.T) 184 | new_covariance = covariance - np.linalg.multi_dot(( 185 | kalman_gain, projected_cov, kalman_gain.T)) 186 | return new_mean, new_covariance 187 | 188 | def gating_distance(self, mean, covariance, measurements, 189 | only_position=False): 190 | """Compute gating distance between state distribution and measurements. 191 | 192 | A suitable distance threshold can be obtained from `chi2inv95`. If 193 | `only_position` is False, the chi-square distribution has 4 degrees of 194 | freedom, otherwise 2. 195 | 196 | Parameters 197 | ---------- 198 | mean : ndarray 199 | Mean vector over the state distribution (8 dimensional). 200 | covariance : ndarray 201 | Covariance of the state distribution (8x8 dimensional). 202 | measurements : ndarray 203 | An Nx4 dimensional matrix of N measurements, each in 204 | format (x, y, a, h) where (x, y) is the bounding box center 205 | position, a the aspect ratio, and h the height. 206 | only_position : Optional[bool] 207 | If True, distance computation is done with respect to the bounding 208 | box center position only. 209 | 210 | Returns 211 | ------- 212 | ndarray 213 | Returns an array of length N, where the i-th element contains the 214 | squared Mahalanobis distance between (mean, covariance) and 215 | `measurements[i]`. 216 | 217 | """ 218 | mean, covariance = self.project(mean, covariance) 219 | if only_position: 220 | mean, covariance = mean[:2], covariance[:2, :2] 221 | measurements = measurements[:, :2] 222 | 223 | cholesky_factor = np.linalg.cholesky(covariance) 224 | d = measurements - mean 225 | z = scipy.linalg.solve_triangular( 226 | cholesky_factor, d.T, lower=True, check_finite=False, 227 | overwrite_b=True) 228 | squared_maha = np.sum(z * z, axis=0) 229 | return squared_maha 230 | -------------------------------------------------------------------------------- /deep_sort/deep_sort/sort/linear_assignment.py: -------------------------------------------------------------------------------- 1 | # vim: expandtab:ts=4:sw=4 2 | from __future__ import absolute_import 3 | import numpy as np 4 | # from sklearn.utils.linear_assignment_ import linear_assignment 5 | from scipy.optimize import linear_sum_assignment as linear_assignment 6 | from . import kalman_filter 7 | 8 | 9 | INFTY_COST = 1e+5 10 | 11 | 12 | def min_cost_matching( 13 | distance_metric, max_distance, tracks, detections, track_indices=None, 14 | detection_indices=None): 15 | """Solve linear assignment problem. 16 | 17 | Parameters 18 | ---------- 19 | distance_metric : Callable[List[Track], List[Detection], List[int], List[int]) -> ndarray 20 | The distance metric is given a list of tracks and detections as well as 21 | a list of N track indices and M detection indices. The metric should 22 | return the NxM dimensional cost matrix, where element (i, j) is the 23 | association cost between the i-th track in the given track indices and 24 | the j-th detection in the given detection_indices. 25 | max_distance : float 26 | Gating threshold. Associations with cost larger than this value are 27 | disregarded. 28 | tracks : List[track.Track] 29 | A list of predicted tracks at the current time step. 30 | detections : List[detection.Detection] 31 | A list of detections at the current time step. 32 | track_indices : List[int] 33 | List of track indices that maps rows in `cost_matrix` to tracks in 34 | `tracks` (see description above). 35 | detection_indices : List[int] 36 | List of detection indices that maps columns in `cost_matrix` to 37 | detections in `detections` (see description above). 38 | 39 | Returns 40 | ------- 41 | (List[(int, int)], List[int], List[int]) 42 | Returns a tuple with the following three entries: 43 | * A list of matched track and detection indices. 44 | * A list of unmatched track indices. 45 | * A list of unmatched detection indices. 46 | 47 | """ 48 | if track_indices is None: 49 | track_indices = np.arange(len(tracks)) 50 | if detection_indices is None: 51 | detection_indices = np.arange(len(detections)) 52 | 53 | if len(detection_indices) == 0 or len(track_indices) == 0: 54 | return [], track_indices, detection_indices # Nothing to match. 55 | 56 | cost_matrix = distance_metric( 57 | tracks, detections, track_indices, detection_indices) 58 | cost_matrix[cost_matrix > max_distance] = max_distance + 1e-5 59 | 60 | row_indices, col_indices = linear_assignment(cost_matrix) 61 | 62 | matches, unmatched_tracks, unmatched_detections = [], [], [] 63 | for col, detection_idx in enumerate(detection_indices): 64 | if col not in col_indices: 65 | unmatched_detections.append(detection_idx) 66 | for row, track_idx in enumerate(track_indices): 67 | if row not in row_indices: 68 | unmatched_tracks.append(track_idx) 69 | for row, col in zip(row_indices, col_indices): 70 | track_idx = track_indices[row] 71 | detection_idx = detection_indices[col] 72 | if cost_matrix[row, col] > max_distance: 73 | unmatched_tracks.append(track_idx) 74 | unmatched_detections.append(detection_idx) 75 | else: 76 | matches.append((track_idx, detection_idx)) 77 | return matches, unmatched_tracks, unmatched_detections 78 | 79 | 80 | def matching_cascade( 81 | distance_metric, max_distance, cascade_depth, tracks, detections, 82 | track_indices=None, detection_indices=None): 83 | """Run matching cascade. 84 | 85 | Parameters 86 | ---------- 87 | distance_metric : Callable[List[Track], List[Detection], List[int], List[int]) -> ndarray 88 | The distance metric is given a list of tracks and detections as well as 89 | a list of N track indices and M detection indices. The metric should 90 | return the NxM dimensional cost matrix, where element (i, j) is the 91 | association cost between the i-th track in the given track indices and 92 | the j-th detection in the given detection indices. 93 | max_distance : float 94 | Gating threshold. Associations with cost larger than this value are 95 | disregarded. 96 | cascade_depth: int 97 | The cascade depth, should be se to the maximum track age. 98 | tracks : List[track.Track] 99 | A list of predicted tracks at the current time step. 100 | detections : List[detection.Detection] 101 | A list of detections at the current time step. 102 | track_indices : Optional[List[int]] 103 | List of track indices that maps rows in `cost_matrix` to tracks in 104 | `tracks` (see description above). Defaults to all tracks. 105 | detection_indices : Optional[List[int]] 106 | List of detection indices that maps columns in `cost_matrix` to 107 | detections in `detections` (see description above). Defaults to all 108 | detections. 109 | 110 | Returns 111 | ------- 112 | (List[(int, int)], List[int], List[int]) 113 | Returns a tuple with the following three entries: 114 | * A list of matched track and detection indices. 115 | * A list of unmatched track indices. 116 | * A list of unmatched detection indices. 117 | 118 | """ 119 | if track_indices is None: 120 | track_indices = list(range(len(tracks))) 121 | if detection_indices is None: 122 | detection_indices = list(range(len(detections))) 123 | 124 | unmatched_detections = detection_indices 125 | matches = [] 126 | for level in range(cascade_depth): 127 | if len(unmatched_detections) == 0: # No detections left 128 | break 129 | 130 | track_indices_l = [ 131 | k for k in track_indices 132 | if tracks[k].time_since_update == 1 + level 133 | ] 134 | if len(track_indices_l) == 0: # Nothing to match at this level 135 | continue 136 | 137 | matches_l, _, unmatched_detections = \ 138 | min_cost_matching( 139 | distance_metric, max_distance, tracks, detections, 140 | track_indices_l, unmatched_detections) 141 | matches += matches_l 142 | unmatched_tracks = list(set(track_indices) - set(k for k, _ in matches)) 143 | return matches, unmatched_tracks, unmatched_detections 144 | 145 | 146 | def gate_cost_matrix( 147 | kf, cost_matrix, tracks, detections, track_indices, detection_indices, 148 | gated_cost=INFTY_COST, only_position=False): 149 | """Invalidate infeasible entries in cost matrix based on the state 150 | distributions obtained by Kalman filtering. 151 | 152 | Parameters 153 | ---------- 154 | kf : The Kalman filter. 155 | cost_matrix : ndarray 156 | The NxM dimensional cost matrix, where N is the number of track indices 157 | and M is the number of detection indices, such that entry (i, j) is the 158 | association cost between `tracks[track_indices[i]]` and 159 | `detections[detection_indices[j]]`. 160 | tracks : List[track.Track] 161 | A list of predicted tracks at the current time step. 162 | detections : List[detection.Detection] 163 | A list of detections at the current time step. 164 | track_indices : List[int] 165 | List of track indices that maps rows in `cost_matrix` to tracks in 166 | `tracks` (see description above). 167 | detection_indices : List[int] 168 | List of detection indices that maps columns in `cost_matrix` to 169 | detections in `detections` (see description above). 170 | gated_cost : Optional[float] 171 | Entries in the cost matrix corresponding to infeasible associations are 172 | set this value. Defaults to a very large value. 173 | only_position : Optional[bool] 174 | If True, only the x, y position of the state distribution is considered 175 | during gating. Defaults to False. 176 | 177 | Returns 178 | ------- 179 | ndarray 180 | Returns the modified cost matrix. 181 | 182 | """ 183 | gating_dim = 2 if only_position else 4 184 | gating_threshold = kalman_filter.chi2inv95[gating_dim] 185 | measurements = np.asarray( 186 | [detections[i].to_xyah() for i in detection_indices]) 187 | for row, track_idx in enumerate(track_indices): 188 | track = tracks[track_idx] 189 | gating_distance = kf.gating_distance( 190 | track.mean, track.covariance, measurements, only_position) 191 | cost_matrix[row, gating_distance > gating_threshold] = gated_cost 192 | return cost_matrix 193 | -------------------------------------------------------------------------------- /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/utils/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/dyh/win10_yolov5_deepsort_counting/adb686ee839e89177990c27f5da35bfae7ab4b9b/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 | 6 | class YamlParser(edict): 7 | """ 8 | This is yaml parser based on EasyDict. 9 | """ 10 | 11 | def __init__(self, cfg_dict=None, config_file=None): 12 | if cfg_dict is None: 13 | cfg_dict = {} 14 | 15 | if config_file is not None: 16 | assert (os.path.isfile(config_file)) 17 | with open(config_file, 'r') as fo: 18 | cfg_dict.update(yaml.safe_load(fo.read())) 19 | 20 | super(YamlParser, self).__init__(cfg_dict) 21 | 22 | def merge_from_file(self, config_file): 23 | with open(config_file, 'r') as fo: 24 | self.update(yaml.safe_load(fo.read())) 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 39 | 40 | ipdb.set_trace() 41 | -------------------------------------------------------------------------------- /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 | -------------------------------------------------------------------------------- /detector.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import numpy as np 3 | 4 | from models.experimental import attempt_load 5 | from utils.datasets import letterbox 6 | from utils.general import non_max_suppression, scale_coords 7 | from utils.torch_utils import select_device 8 | 9 | 10 | class Detector: 11 | 12 | def __init__(self): 13 | self.img_size = 640 14 | self.threshold = 0.3 15 | self.stride = 1 16 | 17 | self.weights = './weights/yolov5m.pt' 18 | 19 | self.device = '0' if torch.cuda.is_available() else 'cpu' 20 | self.device = select_device(self.device) 21 | model = attempt_load(self.weights, map_location=self.device) 22 | model.to(self.device).eval() 23 | model.half() 24 | 25 | self.m = model 26 | self.names = model.module.names if hasattr( 27 | model, 'module') else model.names 28 | 29 | def preprocess(self, img): 30 | 31 | img0 = img.copy() 32 | img = letterbox(img, new_shape=self.img_size)[0] 33 | img = img[:, :, ::-1].transpose(2, 0, 1) 34 | img = np.ascontiguousarray(img) 35 | img = torch.from_numpy(img).to(self.device) 36 | img = img.half() 37 | img /= 255.0 38 | if img.ndimension() == 3: 39 | img = img.unsqueeze(0) 40 | 41 | return img0, img 42 | 43 | def detect(self, im): 44 | 45 | im0, img = self.preprocess(im) 46 | 47 | pred = self.m(img, augment=False)[0] 48 | pred = pred.float() 49 | pred = non_max_suppression(pred, self.threshold, 0.4) 50 | 51 | boxes = [] 52 | for det in pred: 53 | 54 | if det is not None and len(det): 55 | det[:, :4] = scale_coords( 56 | img.shape[2:], det[:, :4], im0.shape).round() 57 | 58 | for *x, conf, cls_id in det: 59 | lbl = self.names[int(cls_id)] 60 | if lbl not in ['person', 'bicycle', 'car', 'motorcycle', 'bus', 'truck']: 61 | continue 62 | pass 63 | x1, y1 = int(x[0]), int(x[1]) 64 | x2, y2 = int(x[2]), int(x[3]) 65 | boxes.append( 66 | (x1, y1, x2, y2, lbl, conf)) 67 | 68 | return boxes 69 | -------------------------------------------------------------------------------- /main.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | 3 | import tracker 4 | from detector import Detector 5 | from cv2 import cv2 6 | 7 | if __name__ == '__main__': 8 | 9 | # 根据视频尺寸，填充一个polygon，供撞线计算使用 10 | mask_image_temp = np.zeros((1080, 1920), dtype=np.uint8) 11 | 12 | # 初始化2个撞线polygon 13 | list_pts_blue = [[204, 305], [227, 431], [605, 522], [1101, 464], [1900, 601], [1902, 495], [1125, 379], [604, 437], 14 | [299, 375], [267, 289]] 15 | ndarray_pts_blue = np.array(list_pts_blue, np.int32) 16 | polygon_blue_value_1 = cv2.fillPoly(mask_image_temp, [ndarray_pts_blue], color=1) 17 | polygon_blue_value_1 = polygon_blue_value_1[:, :, np.newaxis] 18 | 19 | # 填充第二个polygon 20 | mask_image_temp = np.zeros((1080, 1920), dtype=np.uint8) 21 | list_pts_yellow = [[181, 305], [207, 442], [603, 544], [1107, 485], [1898, 625], [1893, 701], [1101, 568], 22 | [594, 637], [118, 483], [109, 303]] 23 | ndarray_pts_yellow = np.array(list_pts_yellow, np.int32) 24 | polygon_yellow_value_2 = cv2.fillPoly(mask_image_temp, [ndarray_pts_yellow], color=2) 25 | polygon_yellow_value_2 = polygon_yellow_value_2[:, :, np.newaxis] 26 | 27 | # 撞线检测用mask，包含2个polygon，（值范围 0、1、2），供撞线计算使用 28 | polygon_mask_blue_and_yellow = polygon_blue_value_1 + polygon_yellow_value_2 29 | 30 | # 缩小尺寸，1920x1080->960x540 31 | polygon_mask_blue_and_yellow = cv2.resize(polygon_mask_blue_and_yellow, (960, 540)) 32 | 33 | # 蓝 色盘 b,g,r 34 | blue_color_plate = [255, 0, 0] 35 | # 蓝 polygon图片 36 | blue_image = np.array(polygon_blue_value_1 * blue_color_plate, np.uint8) 37 | 38 | # 黄 色盘 39 | yellow_color_plate = [0, 255, 255] 40 | # 黄 polygon图片 41 | yellow_image = np.array(polygon_yellow_value_2 * yellow_color_plate, np.uint8) 42 | 43 | # 彩色图片（值范围 0-255） 44 | color_polygons_image = blue_image + yellow_image 45 | # 缩小尺寸，1920x1080->960x540 46 | color_polygons_image = cv2.resize(color_polygons_image, (960, 540)) 47 | 48 | # list 与蓝色polygon重叠 49 | list_overlapping_blue_polygon = [] 50 | 51 | # list 与黄色polygon重叠 52 | list_overlapping_yellow_polygon = [] 53 | 54 | # 进入数量 55 | down_count = 0 56 | # 离开数量 57 | up_count = 0 58 | 59 | font_draw_number = cv2.FONT_HERSHEY_SIMPLEX 60 | draw_text_postion = (int(960 * 0.01), int(540 * 0.05)) 61 | 62 | # 初始化 yolov5 63 | detector = Detector() 64 | 65 | # 打开视频 66 | capture = cv2.VideoCapture(r'video\test.mp4') 67 | # capture = cv2.VideoCapture('TownCentreXVID.avi') 68 | 69 | while True: 70 | # 读取每帧图片 71 | _, im = capture.read() 72 | if im is None: 73 | break 74 | 75 | # 缩小尺寸，1920x1080->960x540 76 | im = cv2.resize(im, (960, 540)) 77 | 78 | list_bboxs = [] 79 | bboxes = detector.detect(im) 80 | 81 | # 如果画面中 有bbox 82 | if len(bboxes) > 0: 83 | list_bboxs = tracker.update(bboxes, im) 84 | 85 | # 画框 86 | # 撞线检测点，(x1，y1)，y方向偏移比例 0.0~1.0 87 | output_image_frame = tracker.draw_bboxes(im, list_bboxs, line_thickness=None) 88 | pass 89 | else: 90 | # 如果画面中 没有bbox 91 | output_image_frame = im 92 | pass 93 | 94 | # 输出图片 95 | output_image_frame = cv2.add(output_image_frame, color_polygons_image) 96 | 97 | if len(list_bboxs) > 0: 98 | # ----------------------判断撞线---------------------- 99 | for item_bbox in list_bboxs: 100 | x1, y1, x2, y2, label, track_id = item_bbox 101 | 102 | # 撞线检测点，(x1，y1)，y方向偏移比例 0.0~1.0 103 | y1_offset = int(y1 + ((y2 - y1) * 0.6)) 104 | 105 | # 撞线的点 106 | y = y1_offset 107 | x = x1 108 | 109 | if polygon_mask_blue_and_yellow[y, x] == 1: 110 | # 如果撞 蓝polygon 111 | if track_id not in list_overlapping_blue_polygon: 112 | list_overlapping_blue_polygon.append(track_id) 113 | pass 114 | 115 | # 判断 黄polygon list 里是否有此 track_id 116 | # 有此 track_id，则 认为是 外出方向 117 | if track_id in list_overlapping_yellow_polygon: 118 | # 外出+1 119 | up_count += 1 120 | 121 | print( 122 | f'类别: {label} | id: {track_id} | 上行撞线 | 上行撞线总数: {up_count} | 上行id列表: {list_overlapping_yellow_polygon}') 123 | 124 | # 删除 黄polygon list 中的此id 125 | list_overlapping_yellow_polygon.remove(track_id) 126 | 127 | pass 128 | else: 129 | # 无此 track_id，不做其他操作 130 | pass 131 | 132 | elif polygon_mask_blue_and_yellow[y, x] == 2: 133 | # 如果撞 黄polygon 134 | if track_id not in list_overlapping_yellow_polygon: 135 | list_overlapping_yellow_polygon.append(track_id) 136 | pass 137 | 138 | # 判断 蓝polygon list 里是否有此 track_id 139 | # 有此 track_id，则 认为是 进入方向 140 | if track_id in list_overlapping_blue_polygon: 141 | # 进入+1 142 | down_count += 1 143 | 144 | print( 145 | f'类别: {label} | id: {track_id} | 下行撞线 | 下行撞线总数: {down_count} | 下行id列表: {list_overlapping_blue_polygon}') 146 | 147 | # 删除 蓝polygon list 中的此id 148 | list_overlapping_blue_polygon.remove(track_id) 149 | 150 | pass 151 | else: 152 | # 无此 track_id，不做其他操作 153 | pass 154 | pass 155 | else: 156 | pass 157 | pass 158 | 159 | pass 160 | 161 | # ----------------------清除无用id---------------------- 162 | list_overlapping_all = list_overlapping_yellow_polygon + list_overlapping_blue_polygon 163 | for id1 in list_overlapping_all: 164 | is_found = False 165 | for _, _, _, _, _, bbox_id in list_bboxs: 166 | if bbox_id == id1: 167 | is_found = True 168 | break 169 | pass 170 | pass 171 | 172 | if not is_found: 173 | # 如果没找到，删除id 174 | if id1 in list_overlapping_yellow_polygon: 175 | list_overlapping_yellow_polygon.remove(id1) 176 | pass 177 | if id1 in list_overlapping_blue_polygon: 178 | list_overlapping_blue_polygon.remove(id1) 179 | pass 180 | pass 181 | list_overlapping_all.clear() 182 | pass 183 | 184 | # 清空list 185 | list_bboxs.clear() 186 | 187 | pass 188 | else: 189 | # 如果图像中没有任何的bbox，则清空list 190 | list_overlapping_blue_polygon.clear() 191 | list_overlapping_yellow_polygon.clear() 192 | pass 193 | pass 194 | 195 | text_draw = 'DOWN: ' + str(down_count) + \ 196 | ' , UP: ' + str(up_count) 197 | output_image_frame = cv2.putText(img=output_image_frame, text=text_draw, 198 | org=draw_text_postion, 199 | fontFace=font_draw_number, 200 | fontScale=1, color=(255, 255, 255), thickness=2) 201 | 202 | cv2.imshow('demo', output_image_frame) 203 | cv2.waitKey(1) 204 | 205 | pass 206 | pass 207 | 208 | capture.release() 209 | cv2.destroyAllWindows() 210 | -------------------------------------------------------------------------------- /models/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/dyh/win10_yolov5_deepsort_counting/adb686ee839e89177990c27f5da35bfae7ab4b9b/models/__init__.py -------------------------------------------------------------------------------- /models/experimental.py: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | """ 3 | Experimental modules 4 | """ 5 | import math 6 | 7 | import numpy as np 8 | import torch 9 | import torch.nn as nn 10 | 11 | from models.common import Conv 12 | from utils.downloads import attempt_download 13 | 14 | 15 | class CrossConv(nn.Module): 16 | # Cross Convolution Downsample 17 | def __init__(self, c1, c2, k=3, s=1, g=1, e=1.0, shortcut=False): 18 | # ch_in, ch_out, kernel, stride, groups, expansion, shortcut 19 | super().__init__() 20 | c_ = int(c2 * e) # hidden channels 21 | self.cv1 = Conv(c1, c_, (1, k), (1, s)) 22 | self.cv2 = Conv(c_, c2, (k, 1), (s, 1), g=g) 23 | self.add = shortcut and c1 == c2 24 | 25 | def forward(self, x): 26 | return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x)) 27 | 28 | 29 | class Sum(nn.Module): 30 | # Weighted sum of 2 or more layers https://arxiv.org/abs/1911.09070 31 | def __init__(self, n, weight=False): # n: number of inputs 32 | super().__init__() 33 | self.weight = weight # apply weights boolean 34 | self.iter = range(n - 1) # iter object 35 | if weight: 36 | self.w = nn.Parameter(-torch.arange(1.0, n) / 2, requires_grad=True) # layer weights 37 | 38 | def forward(self, x): 39 | y = x[0] # no weight 40 | if self.weight: 41 | w = torch.sigmoid(self.w) * 2 42 | for i in self.iter: 43 | y = y + x[i + 1] * w[i] 44 | else: 45 | for i in self.iter: 46 | y = y + x[i + 1] 47 | return y 48 | 49 | 50 | class MixConv2d(nn.Module): 51 | # Mixed Depth-wise Conv https://arxiv.org/abs/1907.09595 52 | def __init__(self, c1, c2, k=(1, 3), s=1, equal_ch=True): # ch_in, ch_out, kernel, stride, ch_strategy 53 | super().__init__() 54 | n = len(k) # number of convolutions 55 | if equal_ch: # equal c_ per group 56 | i = torch.linspace(0, n - 1E-6, c2).floor() # c2 indices 57 | c_ = [(i == g).sum() for g in range(n)] # intermediate channels 58 | else: # equal weight.numel() per group 59 | b = [c2] + [0] * n 60 | a = np.eye(n + 1, n, k=-1) 61 | a -= np.roll(a, 1, axis=1) 62 | a *= np.array(k) ** 2 63 | a[0] = 1 64 | c_ = np.linalg.lstsq(a, b, rcond=None)[0].round() # solve for equal weight indices, ax = b 65 | 66 | self.m = nn.ModuleList( 67 | [nn.Conv2d(c1, int(c_), k, s, k // 2, groups=math.gcd(c1, int(c_)), bias=False) for k, c_ in zip(k, c_)]) 68 | self.bn = nn.BatchNorm2d(c2) 69 | self.act = nn.SiLU() 70 | 71 | def forward(self, x): 72 | return self.act(self.bn(torch.cat([m(x) for m in self.m], 1))) 73 | 74 | 75 | class Ensemble(nn.ModuleList): 76 | # Ensemble of models 77 | def __init__(self): 78 | super().__init__() 79 | 80 | def forward(self, x, augment=False, profile=False, visualize=False): 81 | y = [] 82 | for module in self: 83 | y.append(module(x, augment, profile, visualize)[0]) 84 | # y = torch.stack(y).max(0)[0] # max ensemble 85 | # y = torch.stack(y).mean(0) # mean ensemble 86 | y = torch.cat(y, 1) # nms ensemble 87 | return y, None # inference, train output 88 | 89 | 90 | def attempt_load(weights, map_location=None, inplace=True, fuse=True): 91 | from models.yolo import Detect, Model 92 | 93 | # Loads an ensemble of models weights=[a,b,c] or a single model weights=[a] or weights=a 94 | model = Ensemble() 95 | for w in weights if isinstance(weights, list) else [weights]: 96 | ckpt = torch.load(attempt_download(w), map_location=map_location) # load 97 | if fuse: 98 | model.append(ckpt['ema' if ckpt.get('ema') else 'model'].float().fuse().eval()) # FP32 model 99 | else: 100 | model.append(ckpt['ema' if ckpt.get('ema') else 'model'].float().eval()) # without layer fuse 101 | 102 | # Compatibility updates 103 | for m in model.modules(): 104 | if type(m) in [nn.Hardswish, nn.LeakyReLU, nn.ReLU, nn.ReLU6, nn.SiLU, Detect, Model]: 105 | m.inplace = inplace # pytorch 1.7.0 compatibility 106 | if type(m) is Detect: 107 | if not isinstance(m.anchor_grid, list): # new Detect Layer compatibility 108 | delattr(m, 'anchor_grid') 109 | setattr(m, 'anchor_grid', [torch.zeros(1)] * m.nl) 110 | elif type(m) is Conv: 111 | m._non_persistent_buffers_set = set() # pytorch 1.6.0 compatibility 112 | 113 | if len(model) == 1: 114 | return model[-1] # return model 115 | else: 116 | print(f'Ensemble created with {weights}\n') 117 | for k in ['names']: 118 | setattr(model, k, getattr(model[-1], k)) 119 | model.stride = model[torch.argmax(torch.tensor([m.stride.max() for m in model])).int()].stride # max stride 120 | return model # return ensemble 121 | -------------------------------------------------------------------------------- /models/hub/anchors.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | # Default anchors for COCO data 3 | 4 | 5 | # P5 ------------------------------------------------------------------------------------------------------------------- 6 | # P5-640: 7 | anchors_p5_640: 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 | 13 | # P6 ------------------------------------------------------------------------------------------------------------------- 14 | # P6-640: thr=0.25: 0.9964 BPR, 5.54 anchors past thr, n=12, img_size=640, metric_all=0.281/0.716-mean/best, past_thr=0.469-mean: 9,11, 21,19, 17,41, 43,32, 39,70, 86,64, 65,131, 134,130, 120,265, 282,180, 247,354, 512,387 15 | anchors_p6_640: 16 | - [9,11, 21,19, 17,41] # P3/8 17 | - [43,32, 39,70, 86,64] # P4/16 18 | - [65,131, 134,130, 120,265] # P5/32 19 | - [282,180, 247,354, 512,387] # P6/64 20 | 21 | # P6-1280: thr=0.25: 0.9950 BPR, 5.55 anchors past thr, n=12, img_size=1280, metric_all=0.281/0.714-mean/best, past_thr=0.468-mean: 19,27, 44,40, 38,94, 96,68, 86,152, 180,137, 140,301, 303,264, 238,542, 436,615, 739,380, 925,792 22 | anchors_p6_1280: 23 | - [19,27, 44,40, 38,94] # P3/8 24 | - [96,68, 86,152, 180,137] # P4/16 25 | - [140,301, 303,264, 238,542] # P5/32 26 | - [436,615, 739,380, 925,792] # P6/64 27 | 28 | # P6-1920: thr=0.25: 0.9950 BPR, 5.55 anchors past thr, n=12, img_size=1920, metric_all=0.281/0.714-mean/best, past_thr=0.468-mean: 28,41, 67,59, 57,141, 144,103, 129,227, 270,205, 209,452, 455,396, 358,812, 653,922, 1109,570, 1387,1187 29 | anchors_p6_1920: 30 | - [28,41, 67,59, 57,141] # P3/8 31 | - [144,103, 129,227, 270,205] # P4/16 32 | - [209,452, 455,396, 358,812] # P5/32 33 | - [653,922, 1109,570, 1387,1187] # P6/64 34 | 35 | 36 | # P7 ------------------------------------------------------------------------------------------------------------------- 37 | # P7-640: thr=0.25: 0.9962 BPR, 6.76 anchors past thr, n=15, img_size=640, metric_all=0.275/0.733-mean/best, past_thr=0.466-mean: 11,11, 13,30, 29,20, 30,46, 61,38, 39,92, 78,80, 146,66, 79,163, 149,150, 321,143, 157,303, 257,402, 359,290, 524,372 38 | anchors_p7_640: 39 | - [11,11, 13,30, 29,20] # P3/8 40 | - [30,46, 61,38, 39,92] # P4/16 41 | - [78,80, 146,66, 79,163] # P5/32 42 | - [149,150, 321,143, 157,303] # P6/64 43 | - [257,402, 359,290, 524,372] # P7/128 44 | 45 | # P7-1280: thr=0.25: 0.9968 BPR, 6.71 anchors past thr, n=15, img_size=1280, metric_all=0.273/0.732-mean/best, past_thr=0.463-mean: 19,22, 54,36, 32,77, 70,83, 138,71, 75,173, 165,159, 148,334, 375,151, 334,317, 251,626, 499,474, 750,326, 534,814, 1079,818 46 | anchors_p7_1280: 47 | - [19,22, 54,36, 32,77] # P3/8 48 | - [70,83, 138,71, 75,173] # P4/16 49 | - [165,159, 148,334, 375,151] # P5/32 50 | - [334,317, 251,626, 499,474] # P6/64 51 | - [750,326, 534,814, 1079,818] # P7/128 52 | 53 | # P7-1920: thr=0.25: 0.9968 BPR, 6.71 anchors past thr, n=15, img_size=1920, metric_all=0.273/0.732-mean/best, past_thr=0.463-mean: 29,34, 81,55, 47,115, 105,124, 207,107, 113,259, 247,238, 222,500, 563,227, 501,476, 376,939, 749,711, 1126,489, 801,1222, 1618,1227 54 | anchors_p7_1920: 55 | - [29,34, 81,55, 47,115] # P3/8 56 | - [105,124, 207,107, 113,259] # P4/16 57 | - [247,238, 222,500, 563,227] # P5/32 58 | - [501,476, 376,939, 749,711] # P6/64 59 | - [1126,489, 801,1222, 1618,1227] # P7/128 60 | -------------------------------------------------------------------------------- /models/hub/yolov3-spp.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 1.0 # model depth multiple 6 | width_multiple: 1.0 # layer channel multiple 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 | -------------------------------------------------------------------------------- /models/hub/yolov3-tiny.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 1.0 # model depth multiple 6 | width_multiple: 1.0 # layer channel multiple 7 | anchors: 8 | - [10,14, 23,27, 37,58] # P4/16 9 | - [81,82, 135,169, 344,319] # P5/32 10 | 11 | # YOLOv3-tiny backbone 12 | backbone: 13 | # [from, number, module, args] 14 | [[-1, 1, Conv, [16, 3, 1]], # 0 15 | [-1, 1, nn.MaxPool2d, [2, 2, 0]], # 1-P1/2 16 | [-1, 1, Conv, [32, 3, 1]], 17 | [-1, 1, nn.MaxPool2d, [2, 2, 0]], # 3-P2/4 18 | [-1, 1, Conv, [64, 3, 1]], 19 | [-1, 1, nn.MaxPool2d, [2, 2, 0]], # 5-P3/8 20 | [-1, 1, Conv, [128, 3, 1]], 21 | [-1, 1, nn.MaxPool2d, [2, 2, 0]], # 7-P4/16 22 | [-1, 1, Conv, [256, 3, 1]], 23 | [-1, 1, nn.MaxPool2d, [2, 2, 0]], # 9-P5/32 24 | [-1, 1, Conv, [512, 3, 1]], 25 | [-1, 1, nn.ZeroPad2d, [[0, 1, 0, 1]]], # 11 26 | [-1, 1, nn.MaxPool2d, [2, 1, 0]], # 12 27 | ] 28 | 29 | # YOLOv3-tiny head 30 | head: 31 | [[-1, 1, Conv, [1024, 3, 1]], 32 | [-1, 1, Conv, [256, 1, 1]], 33 | [-1, 1, Conv, [512, 3, 1]], # 15 (P5/32-large) 34 | 35 | [-2, 1, Conv, [128, 1, 1]], 36 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 37 | [[-1, 8], 1, Concat, [1]], # cat backbone P4 38 | [-1, 1, Conv, [256, 3, 1]], # 19 (P4/16-medium) 39 | 40 | [[19, 15], 1, Detect, [nc, anchors]], # Detect(P4, P5) 41 | ] 42 | -------------------------------------------------------------------------------- /models/hub/yolov3.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 1.0 # model depth multiple 6 | width_multiple: 1.0 # layer channel multiple 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 head 29 | head: 30 | [[-1, 1, Bottleneck, [1024, False]], 31 | [-1, 1, Conv, [512, 1, 1]], 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 | -------------------------------------------------------------------------------- /models/hub/yolov5-bifpn.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 1.0 # model depth multiple 6 | width_multiple: 1.0 # layer channel multiple 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 v6.0 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 6, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 3, C3, [1024]], 24 | [-1, 1, SPPF, [1024, 5]], # 9 25 | ] 26 | 27 | # YOLOv5 v6.0 BiFPN 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, C3, [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, C3, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14, 6], 1, Concat, [1]], # cat P4 <--- BiFPN change 41 | [-1, 3, C3, [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, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /models/hub/yolov5-fpn.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 1.0 # model depth multiple 6 | width_multiple: 1.0 # layer channel multiple 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 v6.0 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 6, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 3, C3, [1024]], 24 | [-1, 1, SPPF, [1024, 5]], # 9 25 | ] 26 | 27 | # YOLOv5 v6.0 FPN head 28 | head: 29 | [[-1, 3, C3, [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, C3, [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, C3, [256, False]], # 18 (P3/8-small) 40 | 41 | [[18, 14, 10], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 42 | ] 43 | -------------------------------------------------------------------------------- /models/hub/yolov5-p2.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 1.0 # model depth multiple 6 | width_multiple: 1.0 # layer channel multiple 7 | anchors: 3 # AutoAnchor evolves 3 anchors per P output layer 8 | 9 | # YOLOv5 v6.0 backbone 10 | backbone: 11 | # [from, number, module, args] 12 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 13 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 14 | [-1, 3, C3, [128]], 15 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 16 | [-1, 6, C3, [256]], 17 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 18 | [-1, 9, C3, [512]], 19 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 20 | [-1, 3, C3, [1024]], 21 | [-1, 1, SPPF, [1024, 5]], # 9 22 | ] 23 | 24 | # YOLOv5 v6.0 head with (P2, P3, P4, P5) outputs 25 | head: 26 | [[-1, 1, Conv, [512, 1, 1]], 27 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 28 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 29 | [-1, 3, C3, [512, False]], # 13 30 | 31 | [-1, 1, Conv, [256, 1, 1]], 32 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 33 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 34 | [-1, 3, C3, [256, False]], # 17 (P3/8-small) 35 | 36 | [-1, 1, Conv, [128, 1, 1]], 37 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 38 | [[-1, 2], 1, Concat, [1]], # cat backbone P2 39 | [-1, 1, C3, [128, False]], # 21 (P2/4-xsmall) 40 | 41 | [-1, 1, Conv, [128, 3, 2]], 42 | [[-1, 18], 1, Concat, [1]], # cat head P3 43 | [-1, 3, C3, [256, False]], # 24 (P3/8-small) 44 | 45 | [-1, 1, Conv, [256, 3, 2]], 46 | [[-1, 14], 1, Concat, [1]], # cat head P4 47 | [-1, 3, C3, [512, False]], # 27 (P4/16-medium) 48 | 49 | [-1, 1, Conv, [512, 3, 2]], 50 | [[-1, 10], 1, Concat, [1]], # cat head P5 51 | [-1, 3, C3, [1024, False]], # 30 (P5/32-large) 52 | 53 | [[21, 24, 27, 30], 1, Detect, [nc, anchors]], # Detect(P2, P3, P4, P5) 54 | ] 55 | -------------------------------------------------------------------------------- /models/hub/yolov5-p34.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 0.33 # model depth multiple 6 | width_multiple: 0.50 # layer channel multiple 7 | anchors: 3 # AutoAnchor evolves 3 anchors per P output layer 8 | 9 | # YOLOv5 v6.0 backbone 10 | backbone: 11 | # [from, number, module, args] 12 | [ [ -1, 1, Conv, [ 64, 6, 2, 2 ] ], # 0-P1/2 13 | [ -1, 1, Conv, [ 128, 3, 2 ] ], # 1-P2/4 14 | [ -1, 3, C3, [ 128 ] ], 15 | [ -1, 1, Conv, [ 256, 3, 2 ] ], # 3-P3/8 16 | [ -1, 6, C3, [ 256 ] ], 17 | [ -1, 1, Conv, [ 512, 3, 2 ] ], # 5-P4/16 18 | [ -1, 9, C3, [ 512 ] ], 19 | [ -1, 1, Conv, [ 1024, 3, 2 ] ], # 7-P5/32 20 | [ -1, 3, C3, [ 1024 ] ], 21 | [ -1, 1, SPPF, [ 1024, 5 ] ], # 9 22 | ] 23 | 24 | # YOLOv5 v6.0 head with (P3, P4) outputs 25 | head: 26 | [ [ -1, 1, Conv, [ 512, 1, 1 ] ], 27 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 28 | [ [ -1, 6 ], 1, Concat, [ 1 ] ], # cat backbone P4 29 | [ -1, 3, C3, [ 512, False ] ], # 13 30 | 31 | [ -1, 1, Conv, [ 256, 1, 1 ] ], 32 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 33 | [ [ -1, 4 ], 1, Concat, [ 1 ] ], # cat backbone P3 34 | [ -1, 3, C3, [ 256, False ] ], # 17 (P3/8-small) 35 | 36 | [ -1, 1, Conv, [ 256, 3, 2 ] ], 37 | [ [ -1, 14 ], 1, Concat, [ 1 ] ], # cat head P4 38 | [ -1, 3, C3, [ 512, False ] ], # 20 (P4/16-medium) 39 | 40 | [ [ 17, 20 ], 1, Detect, [ nc, anchors ] ], # Detect(P3, P4) 41 | ] 42 | -------------------------------------------------------------------------------- /models/hub/yolov5-p6.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 1.0 # model depth multiple 6 | width_multiple: 1.0 # layer channel multiple 7 | anchors: 3 # AutoAnchor evolves 3 anchors per P output layer 8 | 9 | # YOLOv5 v6.0 backbone 10 | backbone: 11 | # [from, number, module, args] 12 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 13 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 14 | [-1, 3, C3, [128]], 15 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 16 | [-1, 6, C3, [256]], 17 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 18 | [-1, 9, C3, [512]], 19 | [-1, 1, Conv, [768, 3, 2]], # 7-P5/32 20 | [-1, 3, C3, [768]], 21 | [-1, 1, Conv, [1024, 3, 2]], # 9-P6/64 22 | [-1, 3, C3, [1024]], 23 | [-1, 1, SPPF, [1024, 5]], # 11 24 | ] 25 | 26 | # YOLOv5 v6.0 head with (P3, P4, P5, P6) outputs 27 | head: 28 | [[-1, 1, Conv, [768, 1, 1]], 29 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 30 | [[-1, 8], 1, Concat, [1]], # cat backbone P5 31 | [-1, 3, C3, [768, False]], # 15 32 | 33 | [-1, 1, Conv, [512, 1, 1]], 34 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 35 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 36 | [-1, 3, C3, [512, False]], # 19 37 | 38 | [-1, 1, Conv, [256, 1, 1]], 39 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 40 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 41 | [-1, 3, C3, [256, False]], # 23 (P3/8-small) 42 | 43 | [-1, 1, Conv, [256, 3, 2]], 44 | [[-1, 20], 1, Concat, [1]], # cat head P4 45 | [-1, 3, C3, [512, False]], # 26 (P4/16-medium) 46 | 47 | [-1, 1, Conv, [512, 3, 2]], 48 | [[-1, 16], 1, Concat, [1]], # cat head P5 49 | [-1, 3, C3, [768, False]], # 29 (P5/32-large) 50 | 51 | [-1, 1, Conv, [768, 3, 2]], 52 | [[-1, 12], 1, Concat, [1]], # cat head P6 53 | [-1, 3, C3, [1024, False]], # 32 (P6/64-xlarge) 54 | 55 | [[23, 26, 29, 32], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5, P6) 56 | ] 57 | -------------------------------------------------------------------------------- /models/hub/yolov5-p7.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 1.0 # model depth multiple 6 | width_multiple: 1.0 # layer channel multiple 7 | anchors: 3 # AutoAnchor evolves 3 anchors per P output layer 8 | 9 | # YOLOv5 v6.0 backbone 10 | backbone: 11 | # [from, number, module, args] 12 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 13 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 14 | [-1, 3, C3, [128]], 15 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 16 | [-1, 6, C3, [256]], 17 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 18 | [-1, 9, C3, [512]], 19 | [-1, 1, Conv, [768, 3, 2]], # 7-P5/32 20 | [-1, 3, C3, [768]], 21 | [-1, 1, Conv, [1024, 3, 2]], # 9-P6/64 22 | [-1, 3, C3, [1024]], 23 | [-1, 1, Conv, [1280, 3, 2]], # 11-P7/128 24 | [-1, 3, C3, [1280]], 25 | [-1, 1, SPPF, [1280, 5]], # 13 26 | ] 27 | 28 | # YOLOv5 v6.0 head with (P3, P4, P5, P6, P7) outputs 29 | head: 30 | [[-1, 1, Conv, [1024, 1, 1]], 31 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 32 | [[-1, 10], 1, Concat, [1]], # cat backbone P6 33 | [-1, 3, C3, [1024, False]], # 17 34 | 35 | [-1, 1, Conv, [768, 1, 1]], 36 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 37 | [[-1, 8], 1, Concat, [1]], # cat backbone P5 38 | [-1, 3, C3, [768, False]], # 21 39 | 40 | [-1, 1, Conv, [512, 1, 1]], 41 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 42 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 43 | [-1, 3, C3, [512, False]], # 25 44 | 45 | [-1, 1, Conv, [256, 1, 1]], 46 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 47 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 48 | [-1, 3, C3, [256, False]], # 29 (P3/8-small) 49 | 50 | [-1, 1, Conv, [256, 3, 2]], 51 | [[-1, 26], 1, Concat, [1]], # cat head P4 52 | [-1, 3, C3, [512, False]], # 32 (P4/16-medium) 53 | 54 | [-1, 1, Conv, [512, 3, 2]], 55 | [[-1, 22], 1, Concat, [1]], # cat head P5 56 | [-1, 3, C3, [768, False]], # 35 (P5/32-large) 57 | 58 | [-1, 1, Conv, [768, 3, 2]], 59 | [[-1, 18], 1, Concat, [1]], # cat head P6 60 | [-1, 3, C3, [1024, False]], # 38 (P6/64-xlarge) 61 | 62 | [-1, 1, Conv, [1024, 3, 2]], 63 | [[-1, 14], 1, Concat, [1]], # cat head P7 64 | [-1, 3, C3, [1280, False]], # 41 (P7/128-xxlarge) 65 | 66 | [[29, 32, 35, 38, 41], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5, P6, P7) 67 | ] 68 | -------------------------------------------------------------------------------- /models/hub/yolov5-panet.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 1.0 # model depth multiple 6 | width_multiple: 1.0 # layer channel multiple 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 v6.0 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 6, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 3, C3, [1024]], 24 | [-1, 1, SPPF, [1024, 5]], # 9 25 | ] 26 | 27 | # YOLOv5 v6.0 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, C3, [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, C3, [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, C3, [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, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /models/hub/yolov5l6.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 1.0 # model depth multiple 6 | width_multiple: 1.0 # layer channel multiple 7 | anchors: 8 | - [19,27, 44,40, 38,94] # P3/8 9 | - [96,68, 86,152, 180,137] # P4/16 10 | - [140,301, 303,264, 238,542] # P5/32 11 | - [436,615, 739,380, 925,792] # P6/64 12 | 13 | # YOLOv5 v6.0 backbone 14 | backbone: 15 | # [from, number, module, args] 16 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 17 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 18 | [-1, 3, C3, [128]], 19 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 20 | [-1, 6, C3, [256]], 21 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 22 | [-1, 9, C3, [512]], 23 | [-1, 1, Conv, [768, 3, 2]], # 7-P5/32 24 | [-1, 3, C3, [768]], 25 | [-1, 1, Conv, [1024, 3, 2]], # 9-P6/64 26 | [-1, 3, C3, [1024]], 27 | [-1, 1, SPPF, [1024, 5]], # 11 28 | ] 29 | 30 | # YOLOv5 v6.0 head 31 | head: 32 | [[-1, 1, Conv, [768, 1, 1]], 33 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 34 | [[-1, 8], 1, Concat, [1]], # cat backbone P5 35 | [-1, 3, C3, [768, False]], # 15 36 | 37 | [-1, 1, Conv, [512, 1, 1]], 38 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 39 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 40 | [-1, 3, C3, [512, False]], # 19 41 | 42 | [-1, 1, Conv, [256, 1, 1]], 43 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 44 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 45 | [-1, 3, C3, [256, False]], # 23 (P3/8-small) 46 | 47 | [-1, 1, Conv, [256, 3, 2]], 48 | [[-1, 20], 1, Concat, [1]], # cat head P4 49 | [-1, 3, C3, [512, False]], # 26 (P4/16-medium) 50 | 51 | [-1, 1, Conv, [512, 3, 2]], 52 | [[-1, 16], 1, Concat, [1]], # cat head P5 53 | [-1, 3, C3, [768, False]], # 29 (P5/32-large) 54 | 55 | [-1, 1, Conv, [768, 3, 2]], 56 | [[-1, 12], 1, Concat, [1]], # cat head P6 57 | [-1, 3, C3, [1024, False]], # 32 (P6/64-xlarge) 58 | 59 | [[23, 26, 29, 32], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5, P6) 60 | ] 61 | -------------------------------------------------------------------------------- /models/hub/yolov5m6.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 0.67 # model depth multiple 6 | width_multiple: 0.75 # layer channel multiple 7 | anchors: 8 | - [19,27, 44,40, 38,94] # P3/8 9 | - [96,68, 86,152, 180,137] # P4/16 10 | - [140,301, 303,264, 238,542] # P5/32 11 | - [436,615, 739,380, 925,792] # P6/64 12 | 13 | # YOLOv5 v6.0 backbone 14 | backbone: 15 | # [from, number, module, args] 16 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 17 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 18 | [-1, 3, C3, [128]], 19 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 20 | [-1, 6, C3, [256]], 21 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 22 | [-1, 9, C3, [512]], 23 | [-1, 1, Conv, [768, 3, 2]], # 7-P5/32 24 | [-1, 3, C3, [768]], 25 | [-1, 1, Conv, [1024, 3, 2]], # 9-P6/64 26 | [-1, 3, C3, [1024]], 27 | [-1, 1, SPPF, [1024, 5]], # 11 28 | ] 29 | 30 | # YOLOv5 v6.0 head 31 | head: 32 | [[-1, 1, Conv, [768, 1, 1]], 33 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 34 | [[-1, 8], 1, Concat, [1]], # cat backbone P5 35 | [-1, 3, C3, [768, False]], # 15 36 | 37 | [-1, 1, Conv, [512, 1, 1]], 38 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 39 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 40 | [-1, 3, C3, [512, False]], # 19 41 | 42 | [-1, 1, Conv, [256, 1, 1]], 43 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 44 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 45 | [-1, 3, C3, [256, False]], # 23 (P3/8-small) 46 | 47 | [-1, 1, Conv, [256, 3, 2]], 48 | [[-1, 20], 1, Concat, [1]], # cat head P4 49 | [-1, 3, C3, [512, False]], # 26 (P4/16-medium) 50 | 51 | [-1, 1, Conv, [512, 3, 2]], 52 | [[-1, 16], 1, Concat, [1]], # cat head P5 53 | [-1, 3, C3, [768, False]], # 29 (P5/32-large) 54 | 55 | [-1, 1, Conv, [768, 3, 2]], 56 | [[-1, 12], 1, Concat, [1]], # cat head P6 57 | [-1, 3, C3, [1024, False]], # 32 (P6/64-xlarge) 58 | 59 | [[23, 26, 29, 32], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5, P6) 60 | ] 61 | -------------------------------------------------------------------------------- /models/hub/yolov5n6.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 0.33 # model depth multiple 6 | width_multiple: 0.25 # layer channel multiple 7 | anchors: 8 | - [19,27, 44,40, 38,94] # P3/8 9 | - [96,68, 86,152, 180,137] # P4/16 10 | - [140,301, 303,264, 238,542] # P5/32 11 | - [436,615, 739,380, 925,792] # P6/64 12 | 13 | # YOLOv5 v6.0 backbone 14 | backbone: 15 | # [from, number, module, args] 16 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 17 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 18 | [-1, 3, C3, [128]], 19 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 20 | [-1, 6, C3, [256]], 21 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 22 | [-1, 9, C3, [512]], 23 | [-1, 1, Conv, [768, 3, 2]], # 7-P5/32 24 | [-1, 3, C3, [768]], 25 | [-1, 1, Conv, [1024, 3, 2]], # 9-P6/64 26 | [-1, 3, C3, [1024]], 27 | [-1, 1, SPPF, [1024, 5]], # 11 28 | ] 29 | 30 | # YOLOv5 v6.0 head 31 | head: 32 | [[-1, 1, Conv, [768, 1, 1]], 33 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 34 | [[-1, 8], 1, Concat, [1]], # cat backbone P5 35 | [-1, 3, C3, [768, False]], # 15 36 | 37 | [-1, 1, Conv, [512, 1, 1]], 38 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 39 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 40 | [-1, 3, C3, [512, False]], # 19 41 | 42 | [-1, 1, Conv, [256, 1, 1]], 43 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 44 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 45 | [-1, 3, C3, [256, False]], # 23 (P3/8-small) 46 | 47 | [-1, 1, Conv, [256, 3, 2]], 48 | [[-1, 20], 1, Concat, [1]], # cat head P4 49 | [-1, 3, C3, [512, False]], # 26 (P4/16-medium) 50 | 51 | [-1, 1, Conv, [512, 3, 2]], 52 | [[-1, 16], 1, Concat, [1]], # cat head P5 53 | [-1, 3, C3, [768, False]], # 29 (P5/32-large) 54 | 55 | [-1, 1, Conv, [768, 3, 2]], 56 | [[-1, 12], 1, Concat, [1]], # cat head P6 57 | [-1, 3, C3, [1024, False]], # 32 (P6/64-xlarge) 58 | 59 | [[23, 26, 29, 32], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5, P6) 60 | ] 61 | -------------------------------------------------------------------------------- /models/hub/yolov5s-ghost.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 0.33 # model depth multiple 6 | width_multiple: 0.50 # layer channel multiple 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 v6.0 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 16 | [-1, 1, GhostConv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3Ghost, [128]], 18 | [-1, 1, GhostConv, [256, 3, 2]], # 3-P3/8 19 | [-1, 6, C3Ghost, [256]], 20 | [-1, 1, GhostConv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3Ghost, [512]], 22 | [-1, 1, GhostConv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 3, C3Ghost, [1024]], 24 | [-1, 1, SPPF, [1024, 5]], # 9 25 | ] 26 | 27 | # YOLOv5 v6.0 head 28 | head: 29 | [[-1, 1, GhostConv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, C3Ghost, [512, False]], # 13 33 | 34 | [-1, 1, GhostConv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, C3Ghost, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, GhostConv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, C3Ghost, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, GhostConv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, C3Ghost, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /models/hub/yolov5s-transformer.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 0.33 # model depth multiple 6 | width_multiple: 0.50 # layer channel multiple 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 v6.0 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 6, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 3, C3TR, [1024]], # 9 <--- C3TR() Transformer module 24 | [-1, 1, SPPF, [1024, 5]], # 9 25 | ] 26 | 27 | # YOLOv5 v6.0 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, C3, [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, C3, [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, C3, [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, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /models/hub/yolov5s6.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 0.33 # model depth multiple 6 | width_multiple: 0.50 # layer channel multiple 7 | anchors: 8 | - [19,27, 44,40, 38,94] # P3/8 9 | - [96,68, 86,152, 180,137] # P4/16 10 | - [140,301, 303,264, 238,542] # P5/32 11 | - [436,615, 739,380, 925,792] # P6/64 12 | 13 | # YOLOv5 v6.0 backbone 14 | backbone: 15 | # [from, number, module, args] 16 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 17 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 18 | [-1, 3, C3, [128]], 19 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 20 | [-1, 6, C3, [256]], 21 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 22 | [-1, 9, C3, [512]], 23 | [-1, 1, Conv, [768, 3, 2]], # 7-P5/32 24 | [-1, 3, C3, [768]], 25 | [-1, 1, Conv, [1024, 3, 2]], # 9-P6/64 26 | [-1, 3, C3, [1024]], 27 | [-1, 1, SPPF, [1024, 5]], # 11 28 | ] 29 | 30 | # YOLOv5 v6.0 head 31 | head: 32 | [[-1, 1, Conv, [768, 1, 1]], 33 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 34 | [[-1, 8], 1, Concat, [1]], # cat backbone P5 35 | [-1, 3, C3, [768, False]], # 15 36 | 37 | [-1, 1, Conv, [512, 1, 1]], 38 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 39 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 40 | [-1, 3, C3, [512, False]], # 19 41 | 42 | [-1, 1, Conv, [256, 1, 1]], 43 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 44 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 45 | [-1, 3, C3, [256, False]], # 23 (P3/8-small) 46 | 47 | [-1, 1, Conv, [256, 3, 2]], 48 | [[-1, 20], 1, Concat, [1]], # cat head P4 49 | [-1, 3, C3, [512, False]], # 26 (P4/16-medium) 50 | 51 | [-1, 1, Conv, [512, 3, 2]], 52 | [[-1, 16], 1, Concat, [1]], # cat head P5 53 | [-1, 3, C3, [768, False]], # 29 (P5/32-large) 54 | 55 | [-1, 1, Conv, [768, 3, 2]], 56 | [[-1, 12], 1, Concat, [1]], # cat head P6 57 | [-1, 3, C3, [1024, False]], # 32 (P6/64-xlarge) 58 | 59 | [[23, 26, 29, 32], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5, P6) 60 | ] 61 | -------------------------------------------------------------------------------- /models/hub/yolov5x6.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 1.33 # model depth multiple 6 | width_multiple: 1.25 # layer channel multiple 7 | anchors: 8 | - [19,27, 44,40, 38,94] # P3/8 9 | - [96,68, 86,152, 180,137] # P4/16 10 | - [140,301, 303,264, 238,542] # P5/32 11 | - [436,615, 739,380, 925,792] # P6/64 12 | 13 | # YOLOv5 v6.0 backbone 14 | backbone: 15 | # [from, number, module, args] 16 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 17 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 18 | [-1, 3, C3, [128]], 19 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 20 | [-1, 6, C3, [256]], 21 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 22 | [-1, 9, C3, [512]], 23 | [-1, 1, Conv, [768, 3, 2]], # 7-P5/32 24 | [-1, 3, C3, [768]], 25 | [-1, 1, Conv, [1024, 3, 2]], # 9-P6/64 26 | [-1, 3, C3, [1024]], 27 | [-1, 1, SPPF, [1024, 5]], # 11 28 | ] 29 | 30 | # YOLOv5 v6.0 head 31 | head: 32 | [[-1, 1, Conv, [768, 1, 1]], 33 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 34 | [[-1, 8], 1, Concat, [1]], # cat backbone P5 35 | [-1, 3, C3, [768, False]], # 15 36 | 37 | [-1, 1, Conv, [512, 1, 1]], 38 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 39 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 40 | [-1, 3, C3, [512, False]], # 19 41 | 42 | [-1, 1, Conv, [256, 1, 1]], 43 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 44 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 45 | [-1, 3, C3, [256, False]], # 23 (P3/8-small) 46 | 47 | [-1, 1, Conv, [256, 3, 2]], 48 | [[-1, 20], 1, Concat, [1]], # cat head P4 49 | [-1, 3, C3, [512, False]], # 26 (P4/16-medium) 50 | 51 | [-1, 1, Conv, [512, 3, 2]], 52 | [[-1, 16], 1, Concat, [1]], # cat head P5 53 | [-1, 3, C3, [768, False]], # 29 (P5/32-large) 54 | 55 | [-1, 1, Conv, [768, 3, 2]], 56 | [[-1, 12], 1, Concat, [1]], # cat head P6 57 | [-1, 3, C3, [1024, False]], # 32 (P6/64-xlarge) 58 | 59 | [[23, 26, 29, 32], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5, P6) 60 | ] 61 | -------------------------------------------------------------------------------- /models/yolov5l.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 1.0 # model depth multiple 6 | width_multiple: 1.0 # layer channel multiple 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 v6.0 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 6, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 3, C3, [1024]], 24 | [-1, 1, SPPF, [1024, 5]], # 9 25 | ] 26 | 27 | # YOLOv5 v6.0 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, C3, [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, C3, [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, C3, [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, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /models/yolov5m.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 0.67 # model depth multiple 6 | width_multiple: 0.75 # layer channel multiple 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 v6.0 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 6, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 3, C3, [1024]], 24 | [-1, 1, SPPF, [1024, 5]], # 9 25 | ] 26 | 27 | # YOLOv5 v6.0 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, C3, [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, C3, [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, C3, [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, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /models/yolov5n.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 0.33 # model depth multiple 6 | width_multiple: 0.25 # layer channel multiple 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 v6.0 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 6, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 3, C3, [1024]], 24 | [-1, 1, SPPF, [1024, 5]], # 9 25 | ] 26 | 27 | # YOLOv5 v6.0 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, C3, [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, C3, [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, C3, [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, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /models/yolov5s.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 0.33 # model depth multiple 6 | width_multiple: 0.50 # layer channel multiple 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 v6.0 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 6, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 3, C3, [1024]], 24 | [-1, 1, SPPF, [1024, 5]], # 9 25 | ] 26 | 27 | # YOLOv5 v6.0 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, C3, [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, C3, [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, C3, [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, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /models/yolov5x.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 1.33 # model depth multiple 6 | width_multiple: 1.25 # layer channel multiple 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 v6.0 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 6, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 3, C3, [1024]], 24 | [-1, 1, SPPF, [1024, 5]], # 9 25 | ] 26 | 27 | # YOLOv5 v6.0 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, C3, [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, C3, [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, C3, [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, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /requirements.txt: -------------------------------------------------------------------------------- 1 | # pip install -r requirements.txt 2 | 3 | # Base ---------------------------------------- 4 | matplotlib>=3.2.2 5 | numpy>=1.18.5 6 | opencv-python>=4.1.2 7 | Pillow>=7.1.2 8 | PyYAML>=5.3.1 9 | requests>=2.23.0 10 | scipy>=1.4.1 11 | torch>=1.7.0 12 | torchvision>=0.8.1 13 | tqdm>=4.41.0 14 | 15 | # Logging ------------------------------------- 16 | tensorboard>=2.4.1 17 | # wandb 18 | 19 | # Plotting ------------------------------------ 20 | pandas>=1.1.4 21 | seaborn>=0.11.0 22 | 23 | # Export -------------------------------------- 24 | # coremltools>=4.1 # CoreML export 25 | # onnx>=1.9.0 # ONNX export 26 | # onnx-simplifier>=0.3.6 # ONNX simplifier 27 | # scikit-learn==0.19.2 # CoreML quantization 28 | # tensorflow>=2.4.1 # TFLite export 29 | # tensorflowjs>=3.9.0 # TF.js export 30 | # openvino-dev # OpenVINO export 31 | 32 | # Extras -------------------------------------- 33 | # albumentations>=1.0.3 34 | # Cython # for pycocotools https://github.com/cocodataset/cocoapi/issues/172 35 | # pycocotools>=2.0 # COCO mAP 36 | # roboflow 37 | thop # FLOPs computation 38 | 39 | easydict 40 | -------------------------------------------------------------------------------- /tracker.py: -------------------------------------------------------------------------------- 1 | import cv2 2 | import torch 3 | import numpy as np 4 | 5 | from deep_sort.utils.parser import get_config 6 | from deep_sort.deep_sort import DeepSort 7 | 8 | cfg = get_config() 9 | cfg.merge_from_file("./deep_sort/configs/deep_sort.yaml") 10 | deepsort = DeepSort(cfg.DEEPSORT.REID_CKPT, 11 | max_dist=cfg.DEEPSORT.MAX_DIST, min_confidence=cfg.DEEPSORT.MIN_CONFIDENCE, 12 | nms_max_overlap=cfg.DEEPSORT.NMS_MAX_OVERLAP, max_iou_distance=cfg.DEEPSORT.MAX_IOU_DISTANCE, 13 | max_age=cfg.DEEPSORT.MAX_AGE, n_init=cfg.DEEPSORT.N_INIT, nn_budget=cfg.DEEPSORT.NN_BUDGET, 14 | use_cuda=True) 15 | 16 | 17 | def draw_bboxes(image, bboxes, line_thickness): 18 | line_thickness = line_thickness or round( 19 | 0.002 * (image.shape[0] + image.shape[1]) * 0.5) + 1 20 | 21 | list_pts = [] 22 | point_radius = 4 23 | 24 | for (x1, y1, x2, y2, cls_id, pos_id) in bboxes: 25 | color = (0, 255, 0) 26 | 27 | # 撞线的点 28 | check_point_x = x1 29 | check_point_y = int(y1 + ((y2 - y1) * 0.6)) 30 | 31 | c1, c2 = (x1, y1), (x2, y2) 32 | cv2.rectangle(image, c1, c2, color, thickness=line_thickness, lineType=cv2.LINE_AA) 33 | 34 | font_thickness = max(line_thickness - 1, 1) 35 | t_size = cv2.getTextSize(cls_id, 0, fontScale=line_thickness / 3, thickness=font_thickness)[0] 36 | c2 = c1[0] + t_size[0], c1[1] - t_size[1] - 3 37 | cv2.rectangle(image, c1, c2, color, -1, cv2.LINE_AA) # filled 38 | cv2.putText(image, '{} ID-{}'.format(cls_id, pos_id), (c1[0], c1[1] - 2), 0, line_thickness / 3, 39 | [225, 255, 255], thickness=font_thickness, lineType=cv2.LINE_AA) 40 | 41 | list_pts.append([check_point_x - point_radius, check_point_y - point_radius]) 42 | list_pts.append([check_point_x - point_radius, check_point_y + point_radius]) 43 | list_pts.append([check_point_x + point_radius, check_point_y + point_radius]) 44 | list_pts.append([check_point_x + point_radius, check_point_y - point_radius]) 45 | 46 | ndarray_pts = np.array(list_pts, np.int32) 47 | 48 | cv2.fillPoly(image, [ndarray_pts], color=(0, 0, 255)) 49 | 50 | list_pts.clear() 51 | 52 | return image 53 | 54 | 55 | def update(bboxes, image): 56 | bbox_xywh = [] 57 | confs = [] 58 | bboxes2draw = [] 59 | 60 | if len(bboxes) > 0: 61 | for x1, y1, x2, y2, lbl, conf in bboxes: 62 | obj = [ 63 | int((x1 + x2) * 0.5), int((y1 + y2) * 0.5), 64 | x2 - x1, y2 - y1 65 | ] 66 | bbox_xywh.append(obj) 67 | confs.append(conf) 68 | 69 | xywhs = torch.Tensor(bbox_xywh) 70 | confss = torch.Tensor(confs) 71 | 72 | outputs = deepsort.update(xywhs, confss, image) 73 | 74 | for x1, y1, x2, y2, track_id in list(outputs): 75 | # x1, y1, x2, y2, track_id = value 76 | center_x = (x1 + x2) * 0.5 77 | center_y = (y1 + y2) * 0.5 78 | 79 | label = search_label(center_x=center_x, center_y=center_y, 80 | bboxes_xyxy=bboxes, max_dist_threshold=20.0) 81 | 82 | bboxes2draw.append((x1, y1, x2, y2, label, track_id)) 83 | pass 84 | pass 85 | 86 | return bboxes2draw 87 | 88 | 89 | def search_label(center_x, center_y, bboxes_xyxy, max_dist_threshold): 90 | """ 91 | 在 yolov5 的 bbox 中搜索中心点最接近的label 92 | :param center_x: 93 | :param center_y: 94 | :param bboxes_xyxy: 95 | :param max_dist_threshold: 96 | :return: 字符串 97 | """ 98 | label = '' 99 | # min_label = '' 100 | min_dist = -1.0 101 | 102 | for x1, y1, x2, y2, lbl, conf in bboxes_xyxy: 103 | center_x2 = (x1 + x2) * 0.5 104 | center_y2 = (y1 + y2) * 0.5 105 | 106 | # 横纵距离都小于 max_dist 107 | min_x = abs(center_x2 - center_x) 108 | min_y = abs(center_y2 - center_y) 109 | 110 | if min_x < max_dist_threshold and min_y < max_dist_threshold: 111 | # 距离阈值，判断是否在允许误差范围内 112 | # 取 x, y 方向上的距离平均值 113 | avg_dist = (min_x + min_y) * 0.5 114 | if min_dist == -1.0: 115 | # 第一次赋值 116 | min_dist = avg_dist 117 | # 赋值label 118 | label = lbl 119 | pass 120 | else: 121 | # 若不是第一次，则距离小的优先 122 | if avg_dist < min_dist: 123 | min_dist = avg_dist 124 | # label 125 | label = lbl 126 | pass 127 | pass 128 | pass 129 | 130 | return label 131 | -------------------------------------------------------------------------------- /utils/__init__.py: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | """ 3 | utils/initialization 4 | """ 5 | 6 | 7 | def notebook_init(verbose=True): 8 | # Check system software and hardware 9 | print('Checking setup...') 10 | 11 | import os 12 | import shutil 13 | 14 | from utils.general import check_requirements, emojis, is_colab 15 | from utils.torch_utils import select_device # imports 16 | 17 | check_requirements(('psutil', 'IPython')) 18 | import psutil 19 | from IPython import display # to display images and clear console output 20 | 21 | if is_colab(): 22 | shutil.rmtree('/content/sample_data', ignore_errors=True) # remove colab /sample_data directory 23 | 24 | if verbose: 25 | # System info 26 | # gb = 1 / 1000 ** 3 # bytes to GB 27 | gib = 1 / 1024 ** 3 # bytes to GiB 28 | ram = psutil.virtual_memory().total 29 | total, used, free = shutil.disk_usage("/") 30 | display.clear_output() 31 | s = f'({os.cpu_count()} CPUs, {ram * gib:.1f} GB RAM, {(total - free) * gib:.1f}/{total * gib:.1f} GB disk)' 32 | else: 33 | s = '' 34 | 35 | select_device(newline=False) 36 | print(emojis(f'Setup complete ✅ {s}')) 37 | return display 38 | -------------------------------------------------------------------------------- /utils/activations.py: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | """ 3 | Activation functions 4 | """ 5 | 6 | import torch 7 | import torch.nn as nn 8 | import torch.nn.functional as F 9 | 10 | 11 | # SiLU https://arxiv.org/pdf/1606.08415.pdf ---------------------------------------------------------------------------- 12 | class SiLU(nn.Module): # export-friendly version of nn.SiLU() 13 | @staticmethod 14 | def forward(x): 15 | return x * torch.sigmoid(x) 16 | 17 | 18 | class Hardswish(nn.Module): # export-friendly version of nn.Hardswish() 19 | @staticmethod 20 | def forward(x): 21 | # return x * F.hardsigmoid(x) # for TorchScript and CoreML 22 | return x * F.hardtanh(x + 3, 0.0, 6.0) / 6.0 # for TorchScript, CoreML and ONNX 23 | 24 | 25 | # Mish https://github.com/digantamisra98/Mish -------------------------------------------------------------------------- 26 | class Mish(nn.Module): 27 | @staticmethod 28 | def forward(x): 29 | return x * F.softplus(x).tanh() 30 | 31 | 32 | class MemoryEfficientMish(nn.Module): 33 | class F(torch.autograd.Function): 34 | @staticmethod 35 | def forward(ctx, x): 36 | ctx.save_for_backward(x) 37 | return x.mul(torch.tanh(F.softplus(x))) # x * tanh(ln(1 + exp(x))) 38 | 39 | @staticmethod 40 | def backward(ctx, grad_output): 41 | x = ctx.saved_tensors[0] 42 | sx = torch.sigmoid(x) 43 | fx = F.softplus(x).tanh() 44 | return grad_output * (fx + x * sx * (1 - fx * fx)) 45 | 46 | def forward(self, x): 47 | return self.F.apply(x) 48 | 49 | 50 | # FReLU https://arxiv.org/abs/2007.11824 ------------------------------------------------------------------------------- 51 | class FReLU(nn.Module): 52 | def __init__(self, c1, k=3): # ch_in, kernel 53 | super().__init__() 54 | self.conv = nn.Conv2d(c1, c1, k, 1, 1, groups=c1, bias=False) 55 | self.bn = nn.BatchNorm2d(c1) 56 | 57 | def forward(self, x): 58 | return torch.max(x, self.bn(self.conv(x))) 59 | 60 | 61 | # ACON https://arxiv.org/pdf/2009.04759.pdf ---------------------------------------------------------------------------- 62 | class AconC(nn.Module): 63 | r""" ACON activation (activate or not). 64 | AconC: (p1*x-p2*x) * sigmoid(beta*(p1*x-p2*x)) + p2*x, beta is a learnable parameter 65 | according to "Activate or Not: Learning Customized Activation" . 66 | """ 67 | 68 | def __init__(self, c1): 69 | super().__init__() 70 | self.p1 = nn.Parameter(torch.randn(1, c1, 1, 1)) 71 | self.p2 = nn.Parameter(torch.randn(1, c1, 1, 1)) 72 | self.beta = nn.Parameter(torch.ones(1, c1, 1, 1)) 73 | 74 | def forward(self, x): 75 | dpx = (self.p1 - self.p2) * x 76 | return dpx * torch.sigmoid(self.beta * dpx) + self.p2 * x 77 | 78 | 79 | class MetaAconC(nn.Module): 80 | r""" ACON activation (activate or not). 81 | MetaAconC: (p1*x-p2*x) * sigmoid(beta*(p1*x-p2*x)) + p2*x, beta is generated by a small network 82 | according to "Activate or Not: Learning Customized Activation" . 83 | """ 84 | 85 | def __init__(self, c1, k=1, s=1, r=16): # ch_in, kernel, stride, r 86 | super().__init__() 87 | c2 = max(r, c1 // r) 88 | self.p1 = nn.Parameter(torch.randn(1, c1, 1, 1)) 89 | self.p2 = nn.Parameter(torch.randn(1, c1, 1, 1)) 90 | self.fc1 = nn.Conv2d(c1, c2, k, s, bias=True) 91 | self.fc2 = nn.Conv2d(c2, c1, k, s, bias=True) 92 | # self.bn1 = nn.BatchNorm2d(c2) 93 | # self.bn2 = nn.BatchNorm2d(c1) 94 | 95 | def forward(self, x): 96 | y = x.mean(dim=2, keepdims=True).mean(dim=3, keepdims=True) 97 | # batch-size 1 bug/instabilities https://github.com/ultralytics/yolov5/issues/2891 98 | # beta = torch.sigmoid(self.bn2(self.fc2(self.bn1(self.fc1(y))))) # bug/unstable 99 | beta = torch.sigmoid(self.fc2(self.fc1(y))) # bug patch BN layers removed 100 | dpx = (self.p1 - self.p2) * x 101 | return dpx * torch.sigmoid(beta * dpx) + self.p2 * x 102 | -------------------------------------------------------------------------------- /utils/autoanchor.py: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | """ 3 | AutoAnchor utils 4 | """ 5 | 6 | import random 7 | 8 | import numpy as np 9 | import torch 10 | import yaml 11 | from tqdm import tqdm 12 | 13 | from utils.general import LOGGER, colorstr, emojis 14 | 15 | PREFIX = colorstr('AutoAnchor: ') 16 | 17 | 18 | def check_anchor_order(m): 19 | # Check anchor order against stride order for YOLOv5 Detect() module m, and correct if necessary 20 | a = m.anchors.prod(-1).view(-1) # anchor area 21 | da = a[-1] - a[0] # delta a 22 | ds = m.stride[-1] - m.stride[0] # delta s 23 | if da.sign() != ds.sign(): # same order 24 | LOGGER.info(f'{PREFIX}Reversing anchor order') 25 | m.anchors[:] = m.anchors.flip(0) 26 | 27 | 28 | def check_anchors(dataset, model, thr=4.0, imgsz=640): 29 | # Check anchor fit to data, recompute if necessary 30 | m = model.module.model[-1] if hasattr(model, 'module') else model.model[-1] # Detect() 31 | shapes = imgsz * dataset.shapes / dataset.shapes.max(1, keepdims=True) 32 | scale = np.random.uniform(0.9, 1.1, size=(shapes.shape[0], 1)) # augment scale 33 | wh = torch.tensor(np.concatenate([l[:, 3:5] * s for s, l in zip(shapes * scale, dataset.labels)])).float() # wh 34 | 35 | def metric(k): # compute metric 36 | r = wh[:, None] / k[None] 37 | x = torch.min(r, 1 / r).min(2)[0] # ratio metric 38 | best = x.max(1)[0] # best_x 39 | aat = (x > 1 / thr).float().sum(1).mean() # anchors above threshold 40 | bpr = (best > 1 / thr).float().mean() # best possible recall 41 | return bpr, aat 42 | 43 | anchors = m.anchors.clone() * m.stride.to(m.anchors.device).view(-1, 1, 1) # current anchors 44 | bpr, aat = metric(anchors.cpu().view(-1, 2)) 45 | s = f'\n{PREFIX}{aat:.2f} anchors/target, {bpr:.3f} Best Possible Recall (BPR). ' 46 | if bpr > 0.98: # threshold to recompute 47 | LOGGER.info(emojis(f'{s}Current anchors are a good fit to dataset ✅')) 48 | else: 49 | LOGGER.info(emojis(f'{s}Anchors are a poor fit to dataset ⚠️, attempting to improve...')) 50 | na = m.anchors.numel() // 2 # number of anchors 51 | try: 52 | anchors = kmean_anchors(dataset, n=na, img_size=imgsz, thr=thr, gen=1000, verbose=False) 53 | except Exception as e: 54 | LOGGER.info(f'{PREFIX}ERROR: {e}') 55 | new_bpr = metric(anchors)[0] 56 | if new_bpr > bpr: # replace anchors 57 | anchors = torch.tensor(anchors, device=m.anchors.device).type_as(m.anchors) 58 | m.anchors[:] = anchors.clone().view_as(m.anchors) / m.stride.to(m.anchors.device).view(-1, 1, 1) # loss 59 | check_anchor_order(m) 60 | LOGGER.info(f'{PREFIX}New anchors saved to model. Update model *.yaml to use these anchors in the future.') 61 | else: 62 | LOGGER.info(f'{PREFIX}Original anchors better than new anchors. Proceeding with original anchors.') 63 | 64 | 65 | def kmean_anchors(dataset='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen=1000, verbose=True): 66 | """ Creates kmeans-evolved anchors from training dataset 67 | 68 | Arguments: 69 | dataset: path to data.yaml, or a loaded dataset 70 | n: number of anchors 71 | img_size: image size used for training 72 | thr: anchor-label wh ratio threshold hyperparameter hyp['anchor_t'] used for training, default=4.0 73 | gen: generations to evolve anchors using genetic algorithm 74 | verbose: print all results 75 | 76 | Return: 77 | k: kmeans evolved anchors 78 | 79 | Usage: 80 | from utils.autoanchor import *; _ = kmean_anchors() 81 | """ 82 | from scipy.cluster.vq import kmeans 83 | 84 | npr = np.random 85 | thr = 1 / thr 86 | 87 | def metric(k, wh): # compute metrics 88 | r = wh[:, None] / k[None] 89 | x = torch.min(r, 1 / r).min(2)[0] # ratio metric 90 | # x = wh_iou(wh, torch.tensor(k)) # iou metric 91 | return x, x.max(1)[0] # x, best_x 92 | 93 | def anchor_fitness(k): # mutation fitness 94 | _, best = metric(torch.tensor(k, dtype=torch.float32), wh) 95 | return (best * (best > thr).float()).mean() # fitness 96 | 97 | def print_results(k, verbose=True): 98 | k = k[np.argsort(k.prod(1))] # sort small to large 99 | x, best = metric(k, wh0) 100 | bpr, aat = (best > thr).float().mean(), (x > thr).float().mean() * n # best possible recall, anch > thr 101 | s = f'{PREFIX}thr={thr:.2f}: {bpr:.4f} best possible recall, {aat:.2f} anchors past thr\n' \ 102 | f'{PREFIX}n={n}, img_size={img_size}, metric_all={x.mean():.3f}/{best.mean():.3f}-mean/best, ' \ 103 | f'past_thr={x[x > thr].mean():.3f}-mean: ' 104 | for i, x in enumerate(k): 105 | s += '%i,%i, ' % (round(x[0]), round(x[1])) 106 | if verbose: 107 | LOGGER.info(s[:-2]) 108 | return k 109 | 110 | if isinstance(dataset, str): # *.yaml file 111 | with open(dataset, errors='ignore') as f: 112 | data_dict = yaml.safe_load(f) # model dict 113 | from utils.datasets import LoadImagesAndLabels 114 | dataset = LoadImagesAndLabels(data_dict['train'], augment=True, rect=True) 115 | 116 | # Get label wh 117 | shapes = img_size * dataset.shapes / dataset.shapes.max(1, keepdims=True) 118 | wh0 = np.concatenate([l[:, 3:5] * s for s, l in zip(shapes, dataset.labels)]) # wh 119 | 120 | # Filter 121 | i = (wh0 < 3.0).any(1).sum() 122 | if i: 123 | LOGGER.info(f'{PREFIX}WARNING: Extremely small objects found. {i} of {len(wh0)} labels are < 3 pixels in size.') 124 | wh = wh0[(wh0 >= 2.0).any(1)] # filter > 2 pixels 125 | # wh = wh * (npr.rand(wh.shape[0], 1) * 0.9 + 0.1) # multiply by random scale 0-1 126 | 127 | # Kmeans calculation 128 | LOGGER.info(f'{PREFIX}Running kmeans for {n} anchors on {len(wh)} points...') 129 | s = wh.std(0) # sigmas for whitening 130 | k = kmeans(wh / s, n, iter=30)[0] * s # points 131 | if len(k) != n: # kmeans may return fewer points than requested if wh is insufficient or too similar 132 | LOGGER.warning(f'{PREFIX}WARNING: scipy.cluster.vq.kmeans returned only {len(k)} of {n} requested points') 133 | k = np.sort(npr.rand(n * 2)).reshape(n, 2) * img_size # random init 134 | wh = torch.tensor(wh, dtype=torch.float32) # filtered 135 | wh0 = torch.tensor(wh0, dtype=torch.float32) # unfiltered 136 | k = print_results(k, verbose=False) 137 | 138 | # Plot 139 | # k, d = [None] * 20, [None] * 20 140 | # for i in tqdm(range(1, 21)): 141 | # k[i-1], d[i-1] = kmeans(wh / s, i) # points, mean distance 142 | # fig, ax = plt.subplots(1, 2, figsize=(14, 7), tight_layout=True) 143 | # ax = ax.ravel() 144 | # ax[0].plot(np.arange(1, 21), np.array(d) ** 2, marker='.') 145 | # fig, ax = plt.subplots(1, 2, figsize=(14, 7)) # plot wh 146 | # ax[0].hist(wh[wh[:, 0]<100, 0],400) 147 | # ax[1].hist(wh[wh[:, 1]<100, 1],400) 148 | # fig.savefig('wh.png', dpi=200) 149 | 150 | # Evolve 151 | f, sh, mp, s = anchor_fitness(k), k.shape, 0.9, 0.1 # fitness, generations, mutation prob, sigma 152 | pbar = tqdm(range(gen), desc=f'{PREFIX}Evolving anchors with Genetic Algorithm:') # progress bar 153 | for _ in pbar: 154 | v = np.ones(sh) 155 | while (v == 1).all(): # mutate until a change occurs (prevent duplicates) 156 | v = ((npr.random(sh) < mp) * random.random() * npr.randn(*sh) * s + 1).clip(0.3, 3.0) 157 | kg = (k.copy() * v).clip(min=2.0) 158 | fg = anchor_fitness(kg) 159 | if fg > f: 160 | f, k = fg, kg.copy() 161 | pbar.desc = f'{PREFIX}Evolving anchors with Genetic Algorithm: fitness = {f:.4f}' 162 | if verbose: 163 | print_results(k, verbose) 164 | 165 | return print_results(k) 166 | -------------------------------------------------------------------------------- /utils/autobatch.py: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | """ 3 | Auto-batch utils 4 | """ 5 | 6 | from copy import deepcopy 7 | 8 | import numpy as np 9 | import torch 10 | from torch.cuda import amp 11 | 12 | from utils.general import LOGGER, colorstr 13 | from utils.torch_utils import profile 14 | 15 | 16 | def check_train_batch_size(model, imgsz=640): 17 | # Check YOLOv5 training batch size 18 | with amp.autocast(): 19 | return autobatch(deepcopy(model).train(), imgsz) # compute optimal batch size 20 | 21 | 22 | def autobatch(model, imgsz=640, fraction=0.9, batch_size=16): 23 | # Automatically estimate best batch size to use `fraction` of available CUDA memory 24 | # Usage: 25 | # import torch 26 | # from utils.autobatch import autobatch 27 | # model = torch.hub.load('ultralytics/yolov5', 'yolov5s', autoshape=False) 28 | # print(autobatch(model)) 29 | 30 | prefix = colorstr('AutoBatch: ') 31 | LOGGER.info(f'{prefix}Computing optimal batch size for --imgsz {imgsz}') 32 | device = next(model.parameters()).device # get model device 33 | if device.type == 'cpu': 34 | LOGGER.info(f'{prefix}CUDA not detected, using default CPU batch-size {batch_size}') 35 | return batch_size 36 | 37 | d = str(device).upper() # 'CUDA:0' 38 | properties = torch.cuda.get_device_properties(device) # device properties 39 | t = properties.total_memory / 1024 ** 3 # (GiB) 40 | r = torch.cuda.memory_reserved(device) / 1024 ** 3 # (GiB) 41 | a = torch.cuda.memory_allocated(device) / 1024 ** 3 # (GiB) 42 | f = t - (r + a) # free inside reserved 43 | LOGGER.info(f'{prefix}{d} ({properties.name}) {t:.2f}G total, {r:.2f}G reserved, {a:.2f}G allocated, {f:.2f}G free') 44 | 45 | batch_sizes = [1, 2, 4, 8, 16] 46 | try: 47 | img = [torch.zeros(b, 3, imgsz, imgsz) for b in batch_sizes] 48 | y = profile(img, model, n=3, device=device) 49 | except Exception as e: 50 | LOGGER.warning(f'{prefix}{e}') 51 | 52 | y = [x[2] for x in y if x] # memory [2] 53 | batch_sizes = batch_sizes[:len(y)] 54 | p = np.polyfit(batch_sizes, y, deg=1) # first degree polynomial fit 55 | b = int((f * fraction - p[1]) / p[0]) # y intercept (optimal batch size) 56 | LOGGER.info(f'{prefix}Using batch-size {b} for {d} {t * fraction:.2f}G/{t:.2f}G ({fraction * 100:.0f}%)') 57 | return b 58 | -------------------------------------------------------------------------------- /utils/aws/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/dyh/win10_yolov5_deepsort_counting/adb686ee839e89177990c27f5da35bfae7ab4b9b/utils/aws/__init__.py -------------------------------------------------------------------------------- /utils/aws/mime.sh: -------------------------------------------------------------------------------- 1 | # AWS EC2 instance startup 'MIME' script https://aws.amazon.com/premiumsupport/knowledge-center/execute-user-data-ec2/ 2 | # This script will run on every instance restart, not only on first start 3 | # --- DO NOT COPY ABOVE COMMENTS WHEN PASTING INTO USERDATA --- 4 | 5 | Content-Type: multipart/mixed; boundary="//" 6 | MIME-Version: 1.0 7 | 8 | --// 9 | Content-Type: text/cloud-config; charset="us-ascii" 10 | MIME-Version: 1.0 11 | Content-Transfer-Encoding: 7bit 12 | Content-Disposition: attachment; filename="cloud-config.txt" 13 | 14 | #cloud-config 15 | cloud_final_modules: 16 | - [scripts-user, always] 17 | 18 | --// 19 | Content-Type: text/x-shellscript; charset="us-ascii" 20 | MIME-Version: 1.0 21 | Content-Transfer-Encoding: 7bit 22 | Content-Disposition: attachment; filename="userdata.txt" 23 | 24 | #!/bin/bash 25 | # --- paste contents of userdata.sh here --- 26 | --// 27 | -------------------------------------------------------------------------------- /utils/aws/resume.py: -------------------------------------------------------------------------------- 1 | # Resume all interrupted trainings in yolov5/ dir including DDP trainings 2 | # Usage: $ python utils/aws/resume.py 3 | 4 | import os 5 | import sys 6 | from pathlib import Path 7 | 8 | import torch 9 | import yaml 10 | 11 | FILE = Path(__file__).resolve() 12 | ROOT = FILE.parents[2] # YOLOv5 root directory 13 | if str(ROOT) not in sys.path: 14 | sys.path.append(str(ROOT)) # add ROOT to PATH 15 | 16 | port = 0 # --master_port 17 | path = Path('').resolve() 18 | for last in path.rglob('*/**/last.pt'): 19 | ckpt = torch.load(last) 20 | if ckpt['optimizer'] is None: 21 | continue 22 | 23 | # Load opt.yaml 24 | with open(last.parent.parent / 'opt.yaml', errors='ignore') as f: 25 | opt = yaml.safe_load(f) 26 | 27 | # Get device count 28 | d = opt['device'].split(',') # devices 29 | nd = len(d) # number of devices 30 | ddp = nd > 1 or (nd == 0 and torch.cuda.device_count() > 1) # distributed data parallel 31 | 32 | if ddp: # multi-GPU 33 | port += 1 34 | cmd = f'python -m torch.distributed.run --nproc_per_node {nd} --master_port {port} train.py --resume {last}' 35 | else: # single-GPU 36 | cmd = f'python train.py --resume {last}' 37 | 38 | cmd += ' > /dev/null 2>&1 &' # redirect output to dev/null and run in daemon thread 39 | print(cmd) 40 | os.system(cmd) 41 | -------------------------------------------------------------------------------- /utils/aws/userdata.sh: -------------------------------------------------------------------------------- 1 | #!/bin/bash 2 | # AWS EC2 instance startup script https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/user-data.html 3 | # This script will run only once on first instance start (for a re-start script see mime.sh) 4 | # /home/ubuntu (ubuntu) or /home/ec2-user (amazon-linux) is working dir 5 | # Use >300 GB SSD 6 | 7 | cd home/ubuntu 8 | if [ ! -d yolov5 ]; then 9 | echo "Running first-time script." # install dependencies, download COCO, pull Docker 10 | git clone https://github.com/ultralytics/yolov5 -b master && sudo chmod -R 777 yolov5 11 | cd yolov5 12 | bash data/scripts/get_coco.sh && echo "COCO done." & 13 | sudo docker pull ultralytics/yolov5:latest && echo "Docker done." & 14 | python -m pip install --upgrade pip && pip install -r requirements.txt && python detect.py && echo "Requirements done." & 15 | wait && echo "All tasks done." # finish background tasks 16 | else 17 | echo "Running re-start script." # resume interrupted runs 18 | i=0 19 | list=$(sudo docker ps -qa) # container list i.e. $'one\ntwo\nthree\nfour' 20 | while IFS= read -r id; do 21 | ((i++)) 22 | echo "restarting container $i: $id" 23 | sudo docker start $id 24 | # sudo docker exec -it $id python train.py --resume # single-GPU 25 | sudo docker exec -d $id python utils/aws/resume.py # multi-scenario 26 | done <<<"$list" 27 | fi 28 | -------------------------------------------------------------------------------- /utils/benchmarks.py: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | """ 3 | Run YOLOv5 benchmarks on all supported export formats 4 | 5 | Format | `export.py --include` | Model 6 | --- | --- | --- 7 | PyTorch | - | yolov5s.pt 8 | TorchScript | `torchscript` | yolov5s.torchscript 9 | ONNX | `onnx` | yolov5s.onnx 10 | OpenVINO | `openvino` | yolov5s_openvino_model/ 11 | TensorRT | `engine` | yolov5s.engine 12 | CoreML | `coreml` | yolov5s.mlmodel 13 | TensorFlow SavedModel | `saved_model` | yolov5s_saved_model/ 14 | TensorFlow GraphDef | `pb` | yolov5s.pb 15 | TensorFlow Lite | `tflite` | yolov5s.tflite 16 | TensorFlow Edge TPU | `edgetpu` | yolov5s_edgetpu.tflite 17 | TensorFlow.js | `tfjs` | yolov5s_web_model/ 18 | 19 | Requirements: 20 | $ pip install -r requirements.txt coremltools onnx onnx-simplifier onnxruntime openvino-dev tensorflow-cpu # CPU 21 | $ pip install -r requirements.txt coremltools onnx onnx-simplifier onnxruntime-gpu openvino-dev tensorflow # GPU 22 | 23 | Usage: 24 | $ python utils/benchmarks.py --weights yolov5s.pt --img 640 25 | """ 26 | 27 | import argparse 28 | import sys 29 | import time 30 | from pathlib import Path 31 | 32 | import pandas as pd 33 | 34 | FILE = Path(__file__).resolve() 35 | ROOT = FILE.parents[1] # YOLOv5 root directory 36 | if str(ROOT) not in sys.path: 37 | sys.path.append(str(ROOT)) # add ROOT to PATH 38 | # ROOT = ROOT.relative_to(Path.cwd()) # relative 39 | 40 | import export 41 | import val 42 | from utils import notebook_init 43 | from utils.general import LOGGER, print_args 44 | 45 | 46 | def run(weights=ROOT / 'yolov5s.pt', # weights path 47 | imgsz=640, # inference size (pixels) 48 | batch_size=1, # batch size 49 | data=ROOT / 'data/coco128.yaml', # dataset.yaml path 50 | ): 51 | y, t = [], time.time() 52 | formats = export.export_formats() 53 | for i, (name, f, suffix) in formats.iterrows(): # index, (name, file, suffix) 54 | try: 55 | w = weights if f == '-' else export.run(weights=weights, imgsz=[imgsz], include=[f], device='cpu')[-1] 56 | assert suffix in str(w), 'export failed' 57 | result = val.run(data, w, batch_size, imgsz=imgsz, plots=False, device='cpu', task='benchmark') 58 | metrics = result[0] # metrics (mp, mr, map50, map, *losses(box, obj, cls)) 59 | speeds = result[2] # times (preprocess, inference, postprocess) 60 | y.append([name, metrics[3], speeds[1]]) # mAP, t_inference 61 | except Exception as e: 62 | LOGGER.warning(f'WARNING: Benchmark failure for {name}: {e}') 63 | y.append([name, None, None]) # mAP, t_inference 64 | 65 | # Print results 66 | LOGGER.info('\n') 67 | parse_opt() 68 | notebook_init() # print system info 69 | py = pd.DataFrame(y, columns=['Format', 'mAP@0.5:0.95', 'Inference time (ms)']) 70 | LOGGER.info(f'\nBenchmarks complete ({time.time() - t:.2f}s)') 71 | LOGGER.info(str(py)) 72 | return py 73 | 74 | 75 | def parse_opt(): 76 | parser = argparse.ArgumentParser() 77 | parser.add_argument('--weights', type=str, default=ROOT / 'yolov5s.pt', help='weights path') 78 | parser.add_argument('--imgsz', '--img', '--img-size', type=int, default=640, help='inference size (pixels)') 79 | parser.add_argument('--batch-size', type=int, default=1, help='batch size') 80 | parser.add_argument('--data', type=str, default=ROOT / 'data/coco128.yaml', help='dataset.yaml path') 81 | opt = parser.parse_args() 82 | print_args(FILE.stem, opt) 83 | return opt 84 | 85 | 86 | def main(opt): 87 | run(**vars(opt)) 88 | 89 | 90 | if __name__ == "__main__": 91 | opt = parse_opt() 92 | main(opt) 93 | -------------------------------------------------------------------------------- /utils/callbacks.py: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | """ 3 | Callback utils 4 | """ 5 | 6 | 7 | class Callbacks: 8 | """" 9 | Handles all registered callbacks for YOLOv5 Hooks 10 | """ 11 | 12 | def __init__(self): 13 | # Define the available callbacks 14 | self._callbacks = { 15 | 'on_pretrain_routine_start': [], 16 | 'on_pretrain_routine_end': [], 17 | 18 | 'on_train_start': [], 19 | 'on_train_epoch_start': [], 20 | 'on_train_batch_start': [], 21 | 'optimizer_step': [], 22 | 'on_before_zero_grad': [], 23 | 'on_train_batch_end': [], 24 | 'on_train_epoch_end': [], 25 | 26 | 'on_val_start': [], 27 | 'on_val_batch_start': [], 28 | 'on_val_image_end': [], 29 | 'on_val_batch_end': [], 30 | 'on_val_end': [], 31 | 32 | 'on_fit_epoch_end': [], # fit = train + val 33 | 'on_model_save': [], 34 | 'on_train_end': [], 35 | 'on_params_update': [], 36 | 'teardown': [], 37 | } 38 | self.stop_training = False # set True to interrupt training 39 | 40 | def register_action(self, hook, name='', callback=None): 41 | """ 42 | Register a new action to a callback hook 43 | 44 | Args: 45 | hook The callback hook name to register the action to 46 | name The name of the action for later reference 47 | callback The callback to fire 48 | """ 49 | assert hook in self._callbacks, f"hook '{hook}' not found in callbacks {self._callbacks}" 50 | assert callable(callback), f"callback '{callback}' is not callable" 51 | self._callbacks[hook].append({'name': name, 'callback': callback}) 52 | 53 | def get_registered_actions(self, hook=None): 54 | """" 55 | Returns all the registered actions by callback hook 56 | 57 | Args: 58 | hook The name of the hook to check, defaults to all 59 | """ 60 | if hook: 61 | return self._callbacks[hook] 62 | else: 63 | return self._callbacks 64 | 65 | def run(self, hook, *args, **kwargs): 66 | """ 67 | Loop through the registered actions and fire all callbacks 68 | 69 | Args: 70 | hook The name of the hook to check, defaults to all 71 | args Arguments to receive from YOLOv5 72 | kwargs Keyword Arguments to receive from YOLOv5 73 | """ 74 | 75 | assert hook in self._callbacks, f"hook '{hook}' not found in callbacks {self._callbacks}" 76 | 77 | for logger in self._callbacks[hook]: 78 | logger['callback'](*args, **kwargs) 79 | -------------------------------------------------------------------------------- /utils/downloads.py: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | """ 3 | Download utils 4 | """ 5 | 6 | import os 7 | import platform 8 | import subprocess 9 | import time 10 | import urllib 11 | from pathlib import Path 12 | from zipfile import ZipFile 13 | 14 | import requests 15 | import torch 16 | 17 | 18 | def gsutil_getsize(url=''): 19 | # gs://bucket/file size https://cloud.google.com/storage/docs/gsutil/commands/du 20 | s = subprocess.check_output(f'gsutil du {url}', shell=True).decode('utf-8') 21 | return eval(s.split(' ')[0]) if len(s) else 0 # bytes 22 | 23 | 24 | def safe_download(file, url, url2=None, min_bytes=1E0, error_msg=''): 25 | # Attempts to download file from url or url2, checks and removes incomplete downloads < min_bytes 26 | file = Path(file) 27 | assert_msg = f"Downloaded file '{file}' does not exist or size is < min_bytes={min_bytes}" 28 | try: # url1 29 | print(f'Downloading {url} to {file}...') 30 | torch.hub.download_url_to_file(url, str(file)) 31 | assert file.exists() and file.stat().st_size > min_bytes, assert_msg # check 32 | except Exception as e: # url2 33 | file.unlink(missing_ok=True) # remove partial downloads 34 | print(f'ERROR: {e}\nRe-attempting {url2 or url} to {file}...') 35 | os.system(f"curl -L '{url2 or url}' -o '{file}' --retry 3 -C -") # curl download, retry and resume on fail 36 | finally: 37 | if not file.exists() or file.stat().st_size < min_bytes: # check 38 | file.unlink(missing_ok=True) # remove partial downloads 39 | print(f"ERROR: {assert_msg}\n{error_msg}") 40 | print('') 41 | 42 | 43 | def attempt_download(file, repo='ultralytics/yolov5'): # from utils.downloads import *; attempt_download() 44 | # Attempt file download if does not exist 45 | file = Path(str(file).strip().replace("'", '')) 46 | 47 | if not file.exists(): 48 | # URL specified 49 | name = Path(urllib.parse.unquote(str(file))).name # decode '%2F' to '/' etc. 50 | if str(file).startswith(('http:/', 'https:/')): # download 51 | url = str(file).replace(':/', '://') # Pathlib turns :// -> :/ 52 | file = name.split('?')[0] # parse authentication https://url.com/file.txt?auth... 53 | if Path(file).is_file(): 54 | print(f'Found {url} locally at {file}') # file already exists 55 | else: 56 | safe_download(file=file, url=url, min_bytes=1E5) 57 | return file 58 | 59 | # GitHub assets 60 | file.parent.mkdir(parents=True, exist_ok=True) # make parent dir (if required) 61 | try: 62 | response = requests.get(f'https://api.github.com/repos/{repo}/releases/latest').json() # github api 63 | assets = [x['name'] for x in response['assets']] # release assets, i.e. ['yolov5s.pt', 'yolov5m.pt', ...] 64 | tag = response['tag_name'] # i.e. 'v1.0' 65 | except Exception: # fallback plan 66 | assets = ['yolov5n.pt', 'yolov5s.pt', 'yolov5m.pt', 'yolov5l.pt', 'yolov5x.pt', 67 | 'yolov5n6.pt', 'yolov5s6.pt', 'yolov5m6.pt', 'yolov5l6.pt', 'yolov5x6.pt'] 68 | try: 69 | tag = subprocess.check_output('git tag', shell=True, stderr=subprocess.STDOUT).decode().split()[-1] 70 | except Exception: 71 | tag = 'v6.0' # current release 72 | 73 | if name in assets: 74 | safe_download(file, 75 | url=f'https://github.com/{repo}/releases/download/{tag}/{name}', 76 | # url2=f'https://storage.googleapis.com/{repo}/ckpt/{name}', # backup url (optional) 77 | min_bytes=1E5, 78 | error_msg=f'{file} missing, try downloading from https://github.com/{repo}/releases/') 79 | 80 | return str(file) 81 | 82 | 83 | def gdrive_download(id='16TiPfZj7htmTyhntwcZyEEAejOUxuT6m', file='tmp.zip'): 84 | # Downloads a file from Google Drive. from yolov5.utils.downloads import *; gdrive_download() 85 | t = time.time() 86 | file = Path(file) 87 | cookie = Path('cookie') # gdrive cookie 88 | print(f'Downloading https://drive.google.com/uc?export=download&id={id} as {file}... ', end='') 89 | file.unlink(missing_ok=True) # remove existing file 90 | cookie.unlink(missing_ok=True) # remove existing cookie 91 | 92 | # Attempt file download 93 | out = "NUL" if platform.system() == "Windows" else "/dev/null" 94 | os.system(f'curl -c ./cookie -s -L "drive.google.com/uc?export=download&id={id}" > {out}') 95 | if os.path.exists('cookie'): # large file 96 | s = f'curl -Lb ./cookie "drive.google.com/uc?export=download&confirm={get_token()}&id={id}" -o {file}' 97 | else: # small file 98 | s = f'curl -s -L -o {file} "drive.google.com/uc?export=download&id={id}"' 99 | r = os.system(s) # execute, capture return 100 | cookie.unlink(missing_ok=True) # remove existing cookie 101 | 102 | # Error check 103 | if r != 0: 104 | file.unlink(missing_ok=True) # remove partial 105 | print('Download error ') # raise Exception('Download error') 106 | return r 107 | 108 | # Unzip if archive 109 | if file.suffix == '.zip': 110 | print('unzipping... ', end='') 111 | ZipFile(file).extractall(path=file.parent) # unzip 112 | file.unlink() # remove zip 113 | 114 | print(f'Done ({time.time() - t:.1f}s)') 115 | return r 116 | 117 | 118 | def get_token(cookie="./cookie"): 119 | with open(cookie) as f: 120 | for line in f: 121 | if "download" in line: 122 | return line.split()[-1] 123 | return "" 124 | 125 | # Google utils: https://cloud.google.com/storage/docs/reference/libraries ---------------------------------------------- 126 | # 127 | # 128 | # def upload_blob(bucket_name, source_file_name, destination_blob_name): 129 | # # Uploads a file to a bucket 130 | # # https://cloud.google.com/storage/docs/uploading-objects#storage-upload-object-python 131 | # 132 | # storage_client = storage.Client() 133 | # bucket = storage_client.get_bucket(bucket_name) 134 | # blob = bucket.blob(destination_blob_name) 135 | # 136 | # blob.upload_from_filename(source_file_name) 137 | # 138 | # print('File {} uploaded to {}.'.format( 139 | # source_file_name, 140 | # destination_blob_name)) 141 | # 142 | # 143 | # def download_blob(bucket_name, source_blob_name, destination_file_name): 144 | # # Uploads a blob from a bucket 145 | # storage_client = storage.Client() 146 | # bucket = storage_client.get_bucket(bucket_name) 147 | # blob = bucket.blob(source_blob_name) 148 | # 149 | # blob.download_to_filename(destination_file_name) 150 | # 151 | # print('Blob {} downloaded to {}.'.format( 152 | # source_blob_name, 153 | # destination_file_name)) 154 | -------------------------------------------------------------------------------- /utils/flask_rest_api/README.md: -------------------------------------------------------------------------------- 1 | # Flask REST API 2 | 3 | [REST](https://en.wikipedia.org/wiki/Representational_state_transfer) [API](https://en.wikipedia.org/wiki/API)s are 4 | commonly used to expose Machine Learning (ML) models to other services. This folder contains an example REST API 5 | created using Flask to expose the YOLOv5s model from [PyTorch Hub](https://pytorch.org/hub/ultralytics_yolov5/). 6 | 7 | ## Requirements 8 | 9 | [Flask](https://palletsprojects.com/p/flask/) is required. Install with: 10 | 11 | ```shell 12 | $ pip install Flask 13 | ``` 14 | 15 | ## Run 16 | 17 | After Flask installation run: 18 | 19 | ```shell 20 | $ python3 restapi.py --port 5000 21 | ``` 22 | 23 | Then use [curl](https://curl.se/) to perform a request: 24 | 25 | ```shell 26 | $ curl -X POST -F image=@zidane.jpg 'http://localhost:5000/v1/object-detection/yolov5s' 27 | ``` 28 | 29 | The model inference results are returned as a JSON response: 30 | 31 | ```json 32 | [ 33 | { 34 | "class": 0, 35 | "confidence": 0.8900438547, 36 | "height": 0.9318675399, 37 | "name": "person", 38 | "width": 0.3264600933, 39 | "xcenter": 0.7438579798, 40 | "ycenter": 0.5207948685 41 | }, 42 | { 43 | "class": 0, 44 | "confidence": 0.8440024257, 45 | "height": 0.7155083418, 46 | "name": "person", 47 | "width": 0.6546785235, 48 | "xcenter": 0.427829951, 49 | "ycenter": 0.6334488392 50 | }, 51 | { 52 | "class": 27, 53 | "confidence": 0.3771208823, 54 | "height": 0.3902671337, 55 | "name": "tie", 56 | "width": 0.0696444362, 57 | "xcenter": 0.3675483763, 58 | "ycenter": 0.7991207838 59 | }, 60 | { 61 | "class": 27, 62 | "confidence": 0.3527112305, 63 | "height": 0.1540903747, 64 | "name": "tie", 65 | "width": 0.0336618312, 66 | "xcenter": 0.7814827561, 67 | "ycenter": 0.5065554976 68 | } 69 | ] 70 | ``` 71 | 72 | An example python script to perform inference using [requests](https://docs.python-requests.org/en/master/) is given 73 | in `example_request.py` 74 | -------------------------------------------------------------------------------- /utils/flask_rest_api/example_request.py: -------------------------------------------------------------------------------- 1 | """Perform test request""" 2 | import pprint 3 | 4 | import requests 5 | 6 | DETECTION_URL = "http://localhost:5000/v1/object-detection/yolov5s" 7 | TEST_IMAGE = "zidane.jpg" 8 | 9 | image_data = open(TEST_IMAGE, "rb").read() 10 | 11 | response = requests.post(DETECTION_URL, files={"image": image_data}).json() 12 | 13 | pprint.pprint(response) 14 | -------------------------------------------------------------------------------- /utils/flask_rest_api/restapi.py: -------------------------------------------------------------------------------- 1 | """ 2 | Run a rest API exposing the yolov5s object detection model 3 | """ 4 | import argparse 5 | import io 6 | 7 | import torch 8 | from flask import Flask, request 9 | from PIL import Image 10 | 11 | app = Flask(__name__) 12 | 13 | DETECTION_URL = "/v1/object-detection/yolov5s" 14 | 15 | 16 | @app.route(DETECTION_URL, methods=["POST"]) 17 | def predict(): 18 | if not request.method == "POST": 19 | return 20 | 21 | if request.files.get("image"): 22 | image_file = request.files["image"] 23 | image_bytes = image_file.read() 24 | 25 | img = Image.open(io.BytesIO(image_bytes)) 26 | 27 | results = model(img, size=640) # reduce size=320 for faster inference 28 | return results.pandas().xyxy[0].to_json(orient="records") 29 | 30 | 31 | if __name__ == "__main__": 32 | parser = argparse.ArgumentParser(description="Flask API exposing YOLOv5 model") 33 | parser.add_argument("--port", default=5000, type=int, help="port number") 34 | args = parser.parse_args() 35 | 36 | model = torch.hub.load("ultralytics/yolov5", "yolov5s", force_reload=True) # force_reload to recache 37 | app.run(host="0.0.0.0", port=args.port) # debug=True causes Restarting with stat 38 | -------------------------------------------------------------------------------- /utils/google_app_engine/Dockerfile: -------------------------------------------------------------------------------- 1 | FROM gcr.io/google-appengine/python 2 | 3 | # Create a virtualenv for dependencies. This isolates these packages from 4 | # system-level packages. 5 | # Use -p python3 or -p python3.7 to select python version. Default is version 2. 6 | RUN virtualenv /env -p python3 7 | 8 | # Setting these environment variables are the same as running 9 | # source /env/bin/activate. 10 | ENV VIRTUAL_ENV /env 11 | ENV PATH /env/bin:$PATH 12 | 13 | RUN apt-get update && apt-get install -y python-opencv 14 | 15 | # Copy the application's requirements.txt and run pip to install all 16 | # dependencies into the virtualenv. 17 | ADD requirements.txt /app/requirements.txt 18 | RUN pip install -r /app/requirements.txt 19 | 20 | # Add the application source code. 21 | ADD . /app 22 | 23 | # Run a WSGI server to serve the application. gunicorn must be declared as 24 | # a dependency in requirements.txt. 25 | CMD gunicorn -b :$PORT main:app 26 | -------------------------------------------------------------------------------- /utils/google_app_engine/additional_requirements.txt: -------------------------------------------------------------------------------- 1 | # add these requirements in your app on top of the existing ones 2 | pip==21.1 3 | Flask==1.0.2 4 | gunicorn==19.9.0 5 | -------------------------------------------------------------------------------- /utils/google_app_engine/app.yaml: -------------------------------------------------------------------------------- 1 | runtime: custom 2 | env: flex 3 | 4 | service: yolov5app 5 | 6 | liveness_check: 7 | initial_delay_sec: 600 8 | 9 | manual_scaling: 10 | instances: 1 11 | resources: 12 | cpu: 1 13 | memory_gb: 4 14 | disk_size_gb: 20 15 | -------------------------------------------------------------------------------- /utils/loggers/__init__.py: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | """ 3 | Logging utils 4 | """ 5 | 6 | import os 7 | import warnings 8 | from threading import Thread 9 | 10 | import pkg_resources as pkg 11 | import torch 12 | from torch.utils.tensorboard import SummaryWriter 13 | 14 | from utils.general import colorstr, emojis 15 | from utils.loggers.wandb.wandb_utils import WandbLogger 16 | from utils.plots import plot_images, plot_results 17 | from utils.torch_utils import de_parallel 18 | 19 | LOGGERS = ('csv', 'tb', 'wandb') # text-file, TensorBoard, Weights & Biases 20 | RANK = int(os.getenv('RANK', -1)) 21 | 22 | try: 23 | import wandb 24 | 25 | assert hasattr(wandb, '__version__') # verify package import not local dir 26 | if pkg.parse_version(wandb.__version__) >= pkg.parse_version('0.12.2') and RANK in [0, -1]: 27 | try: 28 | wandb_login_success = wandb.login(timeout=30) 29 | except wandb.errors.UsageError: # known non-TTY terminal issue 30 | wandb_login_success = False 31 | if not wandb_login_success: 32 | wandb = None 33 | except (ImportError, AssertionError): 34 | wandb = None 35 | 36 | 37 | class Loggers(): 38 | # YOLOv5 Loggers class 39 | def __init__(self, save_dir=None, weights=None, opt=None, hyp=None, logger=None, include=LOGGERS): 40 | self.save_dir = save_dir 41 | self.weights = weights 42 | self.opt = opt 43 | self.hyp = hyp 44 | self.logger = logger # for printing results to console 45 | self.include = include 46 | self.keys = ['train/box_loss', 'train/obj_loss', 'train/cls_loss', # train loss 47 | 'metrics/precision', 'metrics/recall', 'metrics/mAP_0.5', 'metrics/mAP_0.5:0.95', # metrics 48 | 'val/box_loss', 'val/obj_loss', 'val/cls_loss', # val loss 49 | 'x/lr0', 'x/lr1', 'x/lr2'] # params 50 | self.best_keys = ['best/epoch', 'best/precision', 'best/recall', 'best/mAP_0.5', 'best/mAP_0.5:0.95',] 51 | for k in LOGGERS: 52 | setattr(self, k, None) # init empty logger dictionary 53 | self.csv = True # always log to csv 54 | 55 | # Message 56 | if not wandb: 57 | prefix = colorstr('Weights & Biases: ') 58 | s = f"{prefix}run 'pip install wandb' to automatically track and visualize YOLOv5 🚀 runs (RECOMMENDED)" 59 | print(emojis(s)) 60 | 61 | # TensorBoard 62 | s = self.save_dir 63 | if 'tb' in self.include and not self.opt.evolve: 64 | prefix = colorstr('TensorBoard: ') 65 | self.logger.info(f"{prefix}Start with 'tensorboard --logdir {s.parent}', view at http://localhost:6006/") 66 | self.tb = SummaryWriter(str(s)) 67 | 68 | # W&B 69 | if wandb and 'wandb' in self.include: 70 | wandb_artifact_resume = isinstance(self.opt.resume, str) and self.opt.resume.startswith('wandb-artifact://') 71 | run_id = torch.load(self.weights).get('wandb_id') if self.opt.resume and not wandb_artifact_resume else None 72 | self.opt.hyp = self.hyp # add hyperparameters 73 | self.wandb = WandbLogger(self.opt, run_id) 74 | else: 75 | self.wandb = None 76 | 77 | def on_pretrain_routine_end(self): 78 | # Callback runs on pre-train routine end 79 | paths = self.save_dir.glob('*labels*.jpg') # training labels 80 | if self.wandb: 81 | self.wandb.log({"Labels": [wandb.Image(str(x), caption=x.name) for x in paths]}) 82 | 83 | def on_train_batch_end(self, ni, model, imgs, targets, paths, plots, sync_bn): 84 | # Callback runs on train batch end 85 | if plots: 86 | if ni == 0: 87 | if not sync_bn: # tb.add_graph() --sync known issue https://github.com/ultralytics/yolov5/issues/3754 88 | with warnings.catch_warnings(): 89 | warnings.simplefilter('ignore') # suppress jit trace warning 90 | self.tb.add_graph(torch.jit.trace(de_parallel(model), imgs[0:1], strict=False), []) 91 | if ni < 3: 92 | f = self.save_dir / f'train_batch{ni}.jpg' # filename 93 | Thread(target=plot_images, args=(imgs, targets, paths, f), daemon=True).start() 94 | if self.wandb and ni == 10: 95 | files = sorted(self.save_dir.glob('train*.jpg')) 96 | self.wandb.log({'Mosaics': [wandb.Image(str(f), caption=f.name) for f in files if f.exists()]}) 97 | 98 | def on_train_epoch_end(self, epoch): 99 | # Callback runs on train epoch end 100 | if self.wandb: 101 | self.wandb.current_epoch = epoch + 1 102 | 103 | def on_val_image_end(self, pred, predn, path, names, im): 104 | # Callback runs on val image end 105 | if self.wandb: 106 | self.wandb.val_one_image(pred, predn, path, names, im) 107 | 108 | def on_val_end(self): 109 | # Callback runs on val end 110 | if self.wandb: 111 | files = sorted(self.save_dir.glob('val*.jpg')) 112 | self.wandb.log({"Validation": [wandb.Image(str(f), caption=f.name) for f in files]}) 113 | 114 | def on_fit_epoch_end(self, vals, epoch, best_fitness, fi): 115 | # Callback runs at the end of each fit (train+val) epoch 116 | x = {k: v for k, v in zip(self.keys, vals)} # dict 117 | if self.csv: 118 | file = self.save_dir / 'results.csv' 119 | n = len(x) + 1 # number of cols 120 | s = '' if file.exists() else (('%20s,' * n % tuple(['epoch'] + self.keys)).rstrip(',') + '\n') # add header 121 | with open(file, 'a') as f: 122 | f.write(s + ('%20.5g,' * n % tuple([epoch] + vals)).rstrip(',') + '\n') 123 | 124 | if self.tb: 125 | for k, v in x.items(): 126 | self.tb.add_scalar(k, v, epoch) 127 | 128 | if self.wandb: 129 | if best_fitness == fi: 130 | best_results = [epoch] + vals[3:7] 131 | for i, name in enumerate(self.best_keys): 132 | self.wandb.wandb_run.summary[name] = best_results[i] # log best results in the summary 133 | self.wandb.log(x) 134 | self.wandb.end_epoch(best_result=best_fitness == fi) 135 | 136 | def on_model_save(self, last, epoch, final_epoch, best_fitness, fi): 137 | # Callback runs on model save event 138 | if self.wandb: 139 | if ((epoch + 1) % self.opt.save_period == 0 and not final_epoch) and self.opt.save_period != -1: 140 | self.wandb.log_model(last.parent, self.opt, epoch, fi, best_model=best_fitness == fi) 141 | 142 | def on_train_end(self, last, best, plots, epoch, results): 143 | # Callback runs on training end 144 | if plots: 145 | plot_results(file=self.save_dir / 'results.csv') # save results.png 146 | files = ['results.png', 'confusion_matrix.png', *(f'{x}_curve.png' for x in ('F1', 'PR', 'P', 'R'))] 147 | files = [(self.save_dir / f) for f in files if (self.save_dir / f).exists()] # filter 148 | 149 | if self.tb: 150 | import cv2 151 | for f in files: 152 | self.tb.add_image(f.stem, cv2.imread(str(f))[..., ::-1], epoch, dataformats='HWC') 153 | 154 | if self.wandb: 155 | self.wandb.log({k: v for k, v in zip(self.keys[3:10], results)}) # log best.pt val results 156 | self.wandb.log({"Results": [wandb.Image(str(f), caption=f.name) for f in files]}) 157 | # Calling wandb.log. TODO: Refactor this into WandbLogger.log_model 158 | if not self.opt.evolve: 159 | wandb.log_artifact(str(best if best.exists() else last), type='model', 160 | name='run_' + self.wandb.wandb_run.id + '_model', 161 | aliases=['latest', 'best', 'stripped']) 162 | self.wandb.finish_run() 163 | 164 | def on_params_update(self, params): 165 | # Update hyperparams or configs of the experiment 166 | # params: A dict containing {param: value} pairs 167 | if self.wandb: 168 | self.wandb.wandb_run.config.update(params, allow_val_change=True) 169 | -------------------------------------------------------------------------------- /utils/loggers/wandb/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/dyh/win10_yolov5_deepsort_counting/adb686ee839e89177990c27f5da35bfae7ab4b9b/utils/loggers/wandb/__init__.py -------------------------------------------------------------------------------- /utils/loggers/wandb/log_dataset.py: -------------------------------------------------------------------------------- 1 | import argparse 2 | 3 | from wandb_utils import WandbLogger 4 | 5 | from utils.general import LOGGER 6 | 7 | WANDB_ARTIFACT_PREFIX = 'wandb-artifact://' 8 | 9 | 10 | def create_dataset_artifact(opt): 11 | logger = WandbLogger(opt, None, job_type='Dataset Creation') # TODO: return value unused 12 | if not logger.wandb: 13 | LOGGER.info("install wandb using `pip install wandb` to log the dataset") 14 | 15 | 16 | if __name__ == '__main__': 17 | parser = argparse.ArgumentParser() 18 | parser.add_argument('--data', type=str, default='data/coco128.yaml', help='data.yaml path') 19 | parser.add_argument('--single-cls', action='store_true', help='train as single-class dataset') 20 | parser.add_argument('--project', type=str, default='YOLOv5', help='name of W&B Project') 21 | parser.add_argument('--entity', default=None, help='W&B entity') 22 | parser.add_argument('--name', type=str, default='log dataset', help='name of W&B run') 23 | 24 | opt = parser.parse_args() 25 | opt.resume = False # Explicitly disallow resume check for dataset upload job 26 | 27 | create_dataset_artifact(opt) 28 | -------------------------------------------------------------------------------- /utils/loggers/wandb/sweep.py: -------------------------------------------------------------------------------- 1 | import sys 2 | from pathlib import Path 3 | 4 | import wandb 5 | 6 | FILE = Path(__file__).resolve() 7 | ROOT = FILE.parents[3] # YOLOv5 root directory 8 | if str(ROOT) not in sys.path: 9 | sys.path.append(str(ROOT)) # add ROOT to PATH 10 | 11 | from train import parse_opt, train 12 | from utils.callbacks import Callbacks 13 | from utils.general import increment_path 14 | from utils.torch_utils import select_device 15 | 16 | 17 | def sweep(): 18 | wandb.init() 19 | # Get hyp dict from sweep agent 20 | hyp_dict = vars(wandb.config).get("_items") 21 | 22 | # Workaround: get necessary opt args 23 | opt = parse_opt(known=True) 24 | opt.batch_size = hyp_dict.get("batch_size") 25 | opt.save_dir = str(increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok or opt.evolve)) 26 | opt.epochs = hyp_dict.get("epochs") 27 | opt.nosave = True 28 | opt.data = hyp_dict.get("data") 29 | opt.weights = str(opt.weights) 30 | opt.cfg = str(opt.cfg) 31 | opt.data = str(opt.data) 32 | opt.hyp = str(opt.hyp) 33 | opt.project = str(opt.project) 34 | device = select_device(opt.device, batch_size=opt.batch_size) 35 | 36 | # train 37 | train(hyp_dict, opt, device, callbacks=Callbacks()) 38 | 39 | 40 | if __name__ == "__main__": 41 | sweep() 42 | -------------------------------------------------------------------------------- /utils/loggers/wandb/sweep.yaml: -------------------------------------------------------------------------------- 1 | # Hyperparameters for training 2 | # To set range- 3 | # Provide min and max values as: 4 | # parameter: 5 | # 6 | # min: scalar 7 | # max: scalar 8 | # OR 9 | # 10 | # Set a specific list of search space- 11 | # parameter: 12 | # values: [scalar1, scalar2, scalar3...] 13 | # 14 | # You can use grid, bayesian and hyperopt search strategy 15 | # For more info on configuring sweeps visit - https://docs.wandb.ai/guides/sweeps/configuration 16 | 17 | program: utils/loggers/wandb/sweep.py 18 | method: random 19 | metric: 20 | name: metrics/mAP_0.5 21 | goal: maximize 22 | 23 | parameters: 24 | # hyperparameters: set either min, max range or values list 25 | data: 26 | value: "data/coco128.yaml" 27 | batch_size: 28 | values: [64] 29 | epochs: 30 | values: [10] 31 | 32 | lr0: 33 | distribution: uniform 34 | min: 1e-5 35 | max: 1e-1 36 | lrf: 37 | distribution: uniform 38 | min: 0.01 39 | max: 1.0 40 | momentum: 41 | distribution: uniform 42 | min: 0.6 43 | max: 0.98 44 | weight_decay: 45 | distribution: uniform 46 | min: 0.0 47 | max: 0.001 48 | warmup_epochs: 49 | distribution: uniform 50 | min: 0.0 51 | max: 5.0 52 | warmup_momentum: 53 | distribution: uniform 54 | min: 0.0 55 | max: 0.95 56 | warmup_bias_lr: 57 | distribution: uniform 58 | min: 0.0 59 | max: 0.2 60 | box: 61 | distribution: uniform 62 | min: 0.02 63 | max: 0.2 64 | cls: 65 | distribution: uniform 66 | min: 0.2 67 | max: 4.0 68 | cls_pw: 69 | distribution: uniform 70 | min: 0.5 71 | max: 2.0 72 | obj: 73 | distribution: uniform 74 | min: 0.2 75 | max: 4.0 76 | obj_pw: 77 | distribution: uniform 78 | min: 0.5 79 | max: 2.0 80 | iou_t: 81 | distribution: uniform 82 | min: 0.1 83 | max: 0.7 84 | anchor_t: 85 | distribution: uniform 86 | min: 2.0 87 | max: 8.0 88 | fl_gamma: 89 | distribution: uniform 90 | min: 0.0 91 | max: 0.1 92 | hsv_h: 93 | distribution: uniform 94 | min: 0.0 95 | max: 0.1 96 | hsv_s: 97 | distribution: uniform 98 | min: 0.0 99 | max: 0.9 100 | hsv_v: 101 | distribution: uniform 102 | min: 0.0 103 | max: 0.9 104 | degrees: 105 | distribution: uniform 106 | min: 0.0 107 | max: 45.0 108 | translate: 109 | distribution: uniform 110 | min: 0.0 111 | max: 0.9 112 | scale: 113 | distribution: uniform 114 | min: 0.0 115 | max: 0.9 116 | shear: 117 | distribution: uniform 118 | min: 0.0 119 | max: 10.0 120 | perspective: 121 | distribution: uniform 122 | min: 0.0 123 | max: 0.001 124 | flipud: 125 | distribution: uniform 126 | min: 0.0 127 | max: 1.0 128 | fliplr: 129 | distribution: uniform 130 | min: 0.0 131 | max: 1.0 132 | mosaic: 133 | distribution: uniform 134 | min: 0.0 135 | max: 1.0 136 | mixup: 137 | distribution: uniform 138 | min: 0.0 139 | max: 1.0 140 | copy_paste: 141 | distribution: uniform 142 | min: 0.0 143 | max: 1.0 144 | -------------------------------------------------------------------------------- /utils/loss.py: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | """ 3 | Loss functions 4 | """ 5 | 6 | import torch 7 | import torch.nn as nn 8 | 9 | from utils.metrics import bbox_iou 10 | from utils.torch_utils import de_parallel 11 | 12 | 13 | def smooth_BCE(eps=0.1): # https://github.com/ultralytics/yolov3/issues/238#issuecomment-598028441 14 | # return positive, negative label smoothing BCE targets 15 | return 1.0 - 0.5 * eps, 0.5 * eps 16 | 17 | 18 | class BCEBlurWithLogitsLoss(nn.Module): 19 | # BCEwithLogitLoss() with reduced missing label effects. 20 | def __init__(self, alpha=0.05): 21 | super().__init__() 22 | self.loss_fcn = nn.BCEWithLogitsLoss(reduction='none') # must be nn.BCEWithLogitsLoss() 23 | self.alpha = alpha 24 | 25 | def forward(self, pred, true): 26 | loss = self.loss_fcn(pred, true) 27 | pred = torch.sigmoid(pred) # prob from logits 28 | dx = pred - true # reduce only missing label effects 29 | # dx = (pred - true).abs() # reduce missing label and false label effects 30 | alpha_factor = 1 - torch.exp((dx - 1) / (self.alpha + 1e-4)) 31 | loss *= alpha_factor 32 | return loss.mean() 33 | 34 | 35 | class FocalLoss(nn.Module): 36 | # Wraps focal loss around existing loss_fcn(), i.e. criteria = FocalLoss(nn.BCEWithLogitsLoss(), gamma=1.5) 37 | def __init__(self, loss_fcn, gamma=1.5, alpha=0.25): 38 | super().__init__() 39 | self.loss_fcn = loss_fcn # must be nn.BCEWithLogitsLoss() 40 | self.gamma = gamma 41 | self.alpha = alpha 42 | self.reduction = loss_fcn.reduction 43 | self.loss_fcn.reduction = 'none' # required to apply FL to each element 44 | 45 | def forward(self, pred, true): 46 | loss = self.loss_fcn(pred, true) 47 | # p_t = torch.exp(-loss) 48 | # loss *= self.alpha * (1.000001 - p_t) ** self.gamma # non-zero power for gradient stability 49 | 50 | # TF implementation https://github.com/tensorflow/addons/blob/v0.7.1/tensorflow_addons/losses/focal_loss.py 51 | pred_prob = torch.sigmoid(pred) # prob from logits 52 | p_t = true * pred_prob + (1 - true) * (1 - pred_prob) 53 | alpha_factor = true * self.alpha + (1 - true) * (1 - self.alpha) 54 | modulating_factor = (1.0 - p_t) ** self.gamma 55 | loss *= alpha_factor * modulating_factor 56 | 57 | if self.reduction == 'mean': 58 | return loss.mean() 59 | elif self.reduction == 'sum': 60 | return loss.sum() 61 | else: # 'none' 62 | return loss 63 | 64 | 65 | class QFocalLoss(nn.Module): 66 | # Wraps Quality focal loss around existing loss_fcn(), i.e. criteria = FocalLoss(nn.BCEWithLogitsLoss(), gamma=1.5) 67 | def __init__(self, loss_fcn, gamma=1.5, alpha=0.25): 68 | super().__init__() 69 | self.loss_fcn = loss_fcn # must be nn.BCEWithLogitsLoss() 70 | self.gamma = gamma 71 | self.alpha = alpha 72 | self.reduction = loss_fcn.reduction 73 | self.loss_fcn.reduction = 'none' # required to apply FL to each element 74 | 75 | def forward(self, pred, true): 76 | loss = self.loss_fcn(pred, true) 77 | 78 | pred_prob = torch.sigmoid(pred) # prob from logits 79 | alpha_factor = true * self.alpha + (1 - true) * (1 - self.alpha) 80 | modulating_factor = torch.abs(true - pred_prob) ** self.gamma 81 | loss *= alpha_factor * modulating_factor 82 | 83 | if self.reduction == 'mean': 84 | return loss.mean() 85 | elif self.reduction == 'sum': 86 | return loss.sum() 87 | else: # 'none' 88 | return loss 89 | 90 | 91 | class ComputeLoss: 92 | # Compute losses 93 | def __init__(self, model, autobalance=False): 94 | self.sort_obj_iou = False 95 | device = next(model.parameters()).device # get model device 96 | h = model.hyp # hyperparameters 97 | 98 | # Define criteria 99 | BCEcls = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['cls_pw']], device=device)) 100 | BCEobj = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['obj_pw']], device=device)) 101 | 102 | # Class label smoothing https://arxiv.org/pdf/1902.04103.pdf eqn 3 103 | self.cp, self.cn = smooth_BCE(eps=h.get('label_smoothing', 0.0)) # positive, negative BCE targets 104 | 105 | # Focal loss 106 | g = h['fl_gamma'] # focal loss gamma 107 | if g > 0: 108 | BCEcls, BCEobj = FocalLoss(BCEcls, g), FocalLoss(BCEobj, g) 109 | 110 | det = de_parallel(model).model[-1] # Detect() module 111 | self.balance = {3: [4.0, 1.0, 0.4]}.get(det.nl, [4.0, 1.0, 0.25, 0.06, 0.02]) # P3-P7 112 | self.ssi = list(det.stride).index(16) if autobalance else 0 # stride 16 index 113 | self.BCEcls, self.BCEobj, self.gr, self.hyp, self.autobalance = BCEcls, BCEobj, 1.0, h, autobalance 114 | for k in 'na', 'nc', 'nl', 'anchors': 115 | setattr(self, k, getattr(det, k)) 116 | 117 | def __call__(self, p, targets): # predictions, targets, model 118 | device = targets.device 119 | lcls, lbox, lobj = torch.zeros(1, device=device), torch.zeros(1, device=device), torch.zeros(1, device=device) 120 | tcls, tbox, indices, anchors = self.build_targets(p, targets) # targets 121 | 122 | # Losses 123 | for i, pi in enumerate(p): # layer index, layer predictions 124 | b, a, gj, gi = indices[i] # image, anchor, gridy, gridx 125 | tobj = torch.zeros_like(pi[..., 0], device=device) # target obj 126 | 127 | n = b.shape[0] # number of targets 128 | if n: 129 | ps = pi[b, a, gj, gi] # prediction subset corresponding to targets 130 | 131 | # Regression 132 | pxy = ps[:, :2].sigmoid() * 2 - 0.5 133 | pwh = (ps[:, 2:4].sigmoid() * 2) ** 2 * anchors[i] 134 | pbox = torch.cat((pxy, pwh), 1) # predicted box 135 | iou = bbox_iou(pbox.T, tbox[i], x1y1x2y2=False, CIoU=True) # iou(prediction, target) 136 | lbox += (1.0 - iou).mean() # iou loss 137 | 138 | # Objectness 139 | score_iou = iou.detach().clamp(0).type(tobj.dtype) 140 | if self.sort_obj_iou: 141 | sort_id = torch.argsort(score_iou) 142 | b, a, gj, gi, score_iou = b[sort_id], a[sort_id], gj[sort_id], gi[sort_id], score_iou[sort_id] 143 | tobj[b, a, gj, gi] = (1.0 - self.gr) + self.gr * score_iou # iou ratio 144 | 145 | # Classification 146 | if self.nc > 1: # cls loss (only if multiple classes) 147 | t = torch.full_like(ps[:, 5:], self.cn, device=device) # targets 148 | t[range(n), tcls[i]] = self.cp 149 | lcls += self.BCEcls(ps[:, 5:], t) # BCE 150 | 151 | # Append targets to text file 152 | # with open('targets.txt', 'a') as file: 153 | # [file.write('%11.5g ' * 4 % tuple(x) + '\n') for x in torch.cat((txy[i], twh[i]), 1)] 154 | 155 | obji = self.BCEobj(pi[..., 4], tobj) 156 | lobj += obji * self.balance[i] # obj loss 157 | if self.autobalance: 158 | self.balance[i] = self.balance[i] * 0.9999 + 0.0001 / obji.detach().item() 159 | 160 | if self.autobalance: 161 | self.balance = [x / self.balance[self.ssi] for x in self.balance] 162 | lbox *= self.hyp['box'] 163 | lobj *= self.hyp['obj'] 164 | lcls *= self.hyp['cls'] 165 | bs = tobj.shape[0] # batch size 166 | 167 | return (lbox + lobj + lcls) * bs, torch.cat((lbox, lobj, lcls)).detach() 168 | 169 | def build_targets(self, p, targets): 170 | # Build targets for compute_loss(), input targets(image,class,x,y,w,h) 171 | na, nt = self.na, targets.shape[0] # number of anchors, targets 172 | tcls, tbox, indices, anch = [], [], [], [] 173 | gain = torch.ones(7, device=targets.device) # normalized to gridspace gain 174 | ai = torch.arange(na, device=targets.device).float().view(na, 1).repeat(1, nt) # same as .repeat_interleave(nt) 175 | targets = torch.cat((targets.repeat(na, 1, 1), ai[:, :, None]), 2) # append anchor indices 176 | 177 | g = 0.5 # bias 178 | off = torch.tensor([[0, 0], 179 | [1, 0], [0, 1], [-1, 0], [0, -1], # j,k,l,m 180 | # [1, 1], [1, -1], [-1, 1], [-1, -1], # jk,jm,lk,lm 181 | ], device=targets.device).float() * g # offsets 182 | 183 | for i in range(self.nl): 184 | anchors = self.anchors[i] 185 | gain[2:6] = torch.tensor(p[i].shape)[[3, 2, 3, 2]] # xyxy gain 186 | 187 | # Match targets to anchors 188 | t = targets * gain 189 | if nt: 190 | # Matches 191 | r = t[:, :, 4:6] / anchors[:, None] # wh ratio 192 | j = torch.max(r, 1 / r).max(2)[0] < self.hyp['anchor_t'] # compare 193 | # j = wh_iou(anchors, t[:, 4:6]) > model.hyp['iou_t'] # iou(3,n)=wh_iou(anchors(3,2), gwh(n,2)) 194 | t = t[j] # filter 195 | 196 | # Offsets 197 | gxy = t[:, 2:4] # grid xy 198 | gxi = gain[[2, 3]] - gxy # inverse 199 | j, k = ((gxy % 1 < g) & (gxy > 1)).T 200 | l, m = ((gxi % 1 < g) & (gxi > 1)).T 201 | j = torch.stack((torch.ones_like(j), j, k, l, m)) 202 | t = t.repeat((5, 1, 1))[j] 203 | offsets = (torch.zeros_like(gxy)[None] + off[:, None])[j] 204 | else: 205 | t = targets[0] 206 | offsets = 0 207 | 208 | # Define 209 | b, c = t[:, :2].long().T # image, class 210 | gxy = t[:, 2:4] # grid xy 211 | gwh = t[:, 4:6] # grid wh 212 | gij = (gxy - offsets).long() 213 | gi, gj = gij.T # grid xy indices 214 | 215 | # Append 216 | a = t[:, 6].long() # anchor indices 217 | indices.append((b, a, gj.clamp_(0, gain[3] - 1), gi.clamp_(0, gain[2] - 1))) # image, anchor, grid indices 218 | tbox.append(torch.cat((gxy - gij, gwh), 1)) # box 219 | anch.append(anchors[a]) # anchors 220 | tcls.append(c) # class 221 | 222 | return tcls, tbox, indices, anch 223 | -------------------------------------------------------------------------------- /video/test.mp4: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/dyh/win10_yolov5_deepsort_counting/adb686ee839e89177990c27f5da35bfae7ab4b9b/video/test.mp4 -------------------------------------------------------------------------------- /weights/yolov5m.pt: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/dyh/win10_yolov5_deepsort_counting/adb686ee839e89177990c27f5da35bfae7ab4b9b/weights/yolov5m.pt --------------------------------------------------------------------------------