├── ObjectDetection_TransferLearning_with_mmdetection
├── configs
│ ├── faster_rcnn_r50_fpn_1x.py
│ ├── faster_rcnn_r50_fpn_1x_our_version.py
│ └── faster_rcnn_x101_64x4d_fpn_1x.py
├── object_detection_transfer_learning_mmdetection.ipynb
└── utils
│ ├── make_annot.py
│ └── viz.py
└── README.md
/ObjectDetection_TransferLearning_with_mmdetection/configs/faster_rcnn_r50_fpn_1x.py:
--------------------------------------------------------------------------------
1 | # model settings
2 | model = dict(
3 | type='FasterRCNN',
4 | pretrained='torchvision://resnet50',
5 | backbone=dict(
6 | type='ResNet',
7 | depth=50,
8 | num_stages=4,
9 | out_indices=(0, 1, 2, 3),
10 | frozen_stages=1,
11 | style='pytorch'),
12 | neck=dict(
13 | type='FPN',
14 | in_channels=[256, 512, 1024, 2048],
15 | out_channels=256,
16 | num_outs=5),
17 | rpn_head=dict(
18 | type='RPNHead',
19 | in_channels=256,
20 | feat_channels=256,
21 | anchor_scales=[8],
22 | anchor_ratios=[0.5, 1.0, 2.0],
23 | anchor_strides=[4, 8, 16, 32, 64],
24 | target_means=[.0, .0, .0, .0],
25 | target_stds=[1.0, 1.0, 1.0, 1.0],
26 | loss_cls=dict(
27 | type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
28 | loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
29 | bbox_roi_extractor=dict(
30 | type='SingleRoIExtractor',
31 | roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
32 | out_channels=256,
33 | featmap_strides=[4, 8, 16, 32]),
34 | bbox_head=dict(
35 | type='SharedFCBBoxHead',
36 | num_fcs=2,
37 | in_channels=256,
38 | fc_out_channels=1024,
39 | roi_feat_size=7,
40 | num_classes=81,
41 | target_means=[0., 0., 0., 0.],
42 | target_stds=[0.1, 0.1, 0.2, 0.2],
43 | reg_class_agnostic=False,
44 | loss_cls=dict(
45 | type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
46 | loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
47 | # model training and testing settings
48 | train_cfg = dict(
49 | rpn=dict(
50 | assigner=dict(
51 | type='MaxIoUAssigner',
52 | pos_iou_thr=0.7,
53 | neg_iou_thr=0.3,
54 | min_pos_iou=0.3,
55 | ignore_iof_thr=-1),
56 | sampler=dict(
57 | type='RandomSampler',
58 | num=256,
59 | pos_fraction=0.5,
60 | neg_pos_ub=-1,
61 | add_gt_as_proposals=False),
62 | allowed_border=0,
63 | pos_weight=-1,
64 | debug=False),
65 | rpn_proposal=dict(
66 | nms_across_levels=False,
67 | nms_pre=2000,
68 | nms_post=2000,
69 | max_num=2000,
70 | nms_thr=0.7,
71 | min_bbox_size=0),
72 | rcnn=dict(
73 | assigner=dict(
74 | type='MaxIoUAssigner',
75 | pos_iou_thr=0.5,
76 | neg_iou_thr=0.5,
77 | min_pos_iou=0.5,
78 | ignore_iof_thr=-1),
79 | sampler=dict(
80 | type='RandomSampler',
81 | num=512,
82 | pos_fraction=0.25,
83 | neg_pos_ub=-1,
84 | add_gt_as_proposals=True),
85 | pos_weight=-1,
86 | debug=False))
87 | test_cfg = dict(
88 | rpn=dict(
89 | nms_across_levels=False,
90 | nms_pre=1000,
91 | nms_post=1000,
92 | max_num=1000,
93 | nms_thr=0.7,
94 | min_bbox_size=0),
95 | rcnn=dict(
96 | score_thr=0.05, nms=dict(type='nms', iou_thr=0.5), max_per_img=100)
97 | # soft-nms is also supported for rcnn testing
98 | # e.g., nms=dict(type='soft_nms', iou_thr=0.5, min_score=0.05)
99 | )
100 | # dataset settings
101 | dataset_type = 'CocoDataset'
102 | data_root = 'data/coco/'
103 | img_norm_cfg = dict(
104 | mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
105 | data = dict(
106 | imgs_per_gpu=2,
107 | workers_per_gpu=2,
108 | train=dict(
109 | type=dataset_type,
110 | ann_file=data_root + 'annotations/instances_train2017.json',
111 | img_prefix=data_root + 'train2017/',
112 | img_scale=(1333, 800),
113 | img_norm_cfg=img_norm_cfg,
114 | size_divisor=32,
115 | flip_ratio=0.5,
116 | with_mask=False,
117 | with_crowd=True,
118 | with_label=True),
119 | val=dict(
120 | type=dataset_type,
121 | ann_file=data_root + 'annotations/instances_val2017.json',
122 | img_prefix=data_root + 'val2017/',
123 | img_scale=(1333, 800),
124 | img_norm_cfg=img_norm_cfg,
125 | size_divisor=32,
126 | flip_ratio=0,
127 | with_mask=False,
128 | with_crowd=True,
129 | with_label=True),
130 | test=dict(
131 | type=dataset_type,
132 | ann_file=data_root + 'annotations/instances_val2017.json',
133 | img_prefix=data_root + 'val2017/',
134 | img_scale=(1333, 800),
135 | img_norm_cfg=img_norm_cfg,
136 | size_divisor=32,
137 | flip_ratio=0,
138 | with_mask=False,
139 | with_label=False,
140 | test_mode=True))
141 | # optimizer
142 | optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
143 | optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
144 | # learning policy
145 | lr_config = dict(
146 | policy='step',
147 | warmup='linear',
148 | warmup_iters=500,
149 | warmup_ratio=1.0 / 3,
150 | step=[8, 11])
151 | checkpoint_config = dict(interval=1)
152 | # yapf:disable
153 | log_config = dict(
154 | interval=50,
155 | hooks=[
156 | dict(type='TextLoggerHook'),
157 | # dict(type='TensorboardLoggerHook')
158 | ])
159 | # yapf:enable
160 | # runtime settings
161 | total_epochs = 12
162 | dist_params = dict(backend='nccl')
163 | log_level = 'INFO'
164 | work_dir = './work_dirs/faster_rcnn_r50_fpn_1x'
165 | load_from = None
166 | resume_from = None
167 | workflow = [('train', 1)]
168 |
--------------------------------------------------------------------------------
/ObjectDetection_TransferLearning_with_mmdetection/configs/faster_rcnn_r50_fpn_1x_our_version.py:
--------------------------------------------------------------------------------
1 | # model settings
2 | model = dict(
3 | type='FasterRCNN',
4 | pretrained='torchvision://resnet50',
5 | backbone=dict(
6 | type='ResNet',
7 | depth=50,
8 | num_stages=4,
9 | out_indices=(0, 1, 2, 3),
10 | frozen_stages=1,
11 | style='pytorch'),
12 | neck=dict(
13 | type='FPN',
14 | in_channels=[256, 512, 1024, 2048],
15 | out_channels=256,
16 | num_outs=5),
17 | rpn_head=dict(
18 | type='RPNHead',
19 | in_channels=256,
20 | feat_channels=256,
21 | anchor_scales=[8],
22 | anchor_ratios=[0.5, 1.0, 2.0],
23 | anchor_strides=[4, 8, 16, 32, 64],
24 | target_means=[.0, .0, .0, .0],
25 | target_stds=[1.0, 1.0, 1.0, 1.0],
26 | loss_cls=dict(
27 | type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
28 | loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
29 | bbox_roi_extractor=dict(
30 | type='SingleRoIExtractor',
31 | roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
32 | out_channels=256,
33 | featmap_strides=[4, 8, 16, 32]),
34 | bbox_head=dict(
35 | type='SharedFCBBoxHead',
36 | num_fcs=2,
37 | in_channels=256,
38 | fc_out_channels=1024,
39 | roi_feat_size=7,
40 | num_classes=5,
41 | target_means=[0., 0., 0., 0.],
42 | target_stds=[0.1, 0.1, 0.2, 0.2],
43 | reg_class_agnostic=False,
44 | loss_cls=dict(
45 | type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
46 | loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
47 | # model training and testing settings
48 | train_cfg = dict(
49 | rpn=dict(
50 | assigner=dict(
51 | type='MaxIoUAssigner',
52 | pos_iou_thr=0.7,
53 | neg_iou_thr=0.3,
54 | min_pos_iou=0.3,
55 | ignore_iof_thr=-1),
56 | sampler=dict(
57 | type='RandomSampler',
58 | num=256,
59 | pos_fraction=0.5,
60 | neg_pos_ub=-1,
61 | add_gt_as_proposals=False),
62 | allowed_border=0,
63 | pos_weight=-1,
64 | debug=False),
65 | rpn_proposal=dict(
66 | nms_across_levels=False,
67 | nms_pre=2000,
68 | nms_post=2000,
69 | max_num=2000,
70 | nms_thr=0.7,
71 | min_bbox_size=0),
72 | rcnn=dict(
73 | assigner=dict(
74 | type='MaxIoUAssigner',
75 | pos_iou_thr=0.5,
76 | neg_iou_thr=0.5,
77 | min_pos_iou=0.5,
78 | ignore_iof_thr=-1),
79 | sampler=dict(
80 | type='RandomSampler',
81 | num=512,
82 | pos_fraction=0.25,
83 | neg_pos_ub=-1,
84 | add_gt_as_proposals=True),
85 | pos_weight=-1,
86 | debug=False))
87 | test_cfg = dict(
88 | rpn=dict(
89 | nms_across_levels=False,
90 | nms_pre=1000,
91 | nms_post=1000,
92 | max_num=1000,
93 | nms_thr=0.7,
94 | min_bbox_size=0),
95 | rcnn=dict(
96 | score_thr=0.05, nms=dict(type='nms', iou_thr=0.5), max_per_img=100)
97 | # soft-nms is also supported for rcnn testing
98 | # e.g., nms=dict(type='soft_nms', iou_thr=0.5, min_score=0.05)
99 | )
100 | # dataset settings
101 | dataset_type = 'CustomDataset'
102 | data_root = ''
103 | img_norm_cfg = dict(
104 | mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
105 | data = dict(
106 | imgs_per_gpu=6,
107 | workers_per_gpu=6,
108 | train=dict(
109 | type=dataset_type,
110 | ann_file='annotations_train.pkl',
111 | img_prefix='',
112 | img_scale=(1000, 650),
113 | img_norm_cfg=img_norm_cfg,
114 | size_divisor=32,
115 | flip_ratio=0.5,
116 | with_mask=False,
117 | with_crowd=True,
118 | with_label=True),
119 | val=dict(
120 | type=dataset_type,
121 | ann_file='annotations_val.pkl',
122 | img_prefix='',
123 | img_scale=(1000, 650),
124 | img_norm_cfg=img_norm_cfg,
125 | size_divisor=32,
126 | flip_ratio=0,
127 | with_mask=False,
128 | with_crowd=True,
129 | with_label=True),
130 | test=dict(
131 | type=dataset_type,
132 | ann_file=None,
133 | img_prefix='',
134 | img_scale=(1000, 650),
135 | img_norm_cfg=img_norm_cfg,
136 | size_divisor=32,
137 | flip_ratio=0,
138 | with_mask=False,
139 | with_label=False,
140 | test_mode=True))
141 | # optimizer
142 | optimizer = dict(type='SGD', lr=0.0025, momentum=0.9, weight_decay=0.0001)
143 | optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
144 | # learning policy
145 | lr_config = dict(
146 | policy='step',
147 | warmup='linear',
148 | warmup_iters=500,
149 | warmup_ratio=1.0 / 3,
150 | step=[6, 9])
151 | checkpoint_config = dict(interval=1)
152 | # yapf:disable
153 | log_config = dict(
154 | interval=60,
155 | hooks=[
156 | dict(type='TextLoggerHook'),
157 | # dict(type='TensorboardLoggerHook')
158 | ])
159 | # yapf:enable
160 | # runtime settings
161 | total_epochs = 12
162 | dist_params = dict(backend='nccl')
163 | log_level = 'INFO'
164 | work_dir = './work_dirs/faster_rcnn_r50_fpn_1x'
165 | load_from = None
166 | resume_from = None
167 | workflow = [('train', 1)]
168 |
--------------------------------------------------------------------------------
/ObjectDetection_TransferLearning_with_mmdetection/configs/faster_rcnn_x101_64x4d_fpn_1x.py:
--------------------------------------------------------------------------------
1 | # model settings
2 | model = dict(
3 | type='FasterRCNN',
4 | pretrained='open-mmlab://resnext101_64x4d',
5 | backbone=dict(
6 | type='ResNeXt',
7 | depth=101,
8 | groups=64,
9 | base_width=4,
10 | num_stages=4,
11 | out_indices=(0, 1, 2, 3),
12 | frozen_stages=1,
13 | style='pytorch'),
14 | neck=dict(
15 | type='FPN',
16 | in_channels=[256, 512, 1024, 2048],
17 | out_channels=256,
18 | num_outs=5),
19 | rpn_head=dict(
20 | type='RPNHead',
21 | in_channels=256,
22 | feat_channels=256,
23 | anchor_scales=[8],
24 | anchor_ratios=[0.5, 1.0, 2.0],
25 | anchor_strides=[4, 8, 16, 32, 64],
26 | target_means=[.0, .0, .0, .0],
27 | target_stds=[1.0, 1.0, 1.0, 1.0],
28 | loss_cls=dict(
29 | type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
30 | loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
31 | bbox_roi_extractor=dict(
32 | type='SingleRoIExtractor',
33 | roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
34 | out_channels=256,
35 | featmap_strides=[4, 8, 16, 32]),
36 | bbox_head=dict(
37 | type='SharedFCBBoxHead',
38 | num_fcs=2,
39 | in_channels=256,
40 | fc_out_channels=1024,
41 | roi_feat_size=7,
42 | num_classes=81,
43 | target_means=[0., 0., 0., 0.],
44 | target_stds=[0.1, 0.1, 0.2, 0.2],
45 | reg_class_agnostic=False,
46 | loss_cls=dict(
47 | type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
48 | loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
49 | # model training and testing settings
50 | train_cfg = dict(
51 | rpn=dict(
52 | assigner=dict(
53 | type='MaxIoUAssigner',
54 | pos_iou_thr=0.7,
55 | neg_iou_thr=0.3,
56 | min_pos_iou=0.3,
57 | ignore_iof_thr=-1),
58 | sampler=dict(
59 | type='RandomSampler',
60 | num=256,
61 | pos_fraction=0.5,
62 | neg_pos_ub=-1,
63 | add_gt_as_proposals=False),
64 | allowed_border=0,
65 | pos_weight=-1,
66 | debug=False),
67 | rpn_proposal=dict(
68 | nms_across_levels=False,
69 | nms_pre=2000,
70 | nms_post=2000,
71 | max_num=2000,
72 | nms_thr=0.7,
73 | min_bbox_size=0),
74 | rcnn=dict(
75 | assigner=dict(
76 | type='MaxIoUAssigner',
77 | pos_iou_thr=0.5,
78 | neg_iou_thr=0.5,
79 | min_pos_iou=0.5,
80 | ignore_iof_thr=-1),
81 | sampler=dict(
82 | type='RandomSampler',
83 | num=512,
84 | pos_fraction=0.25,
85 | neg_pos_ub=-1,
86 | add_gt_as_proposals=True),
87 | pos_weight=-1,
88 | debug=False))
89 | test_cfg = dict(
90 | rpn=dict(
91 | nms_across_levels=False,
92 | nms_pre=1000,
93 | nms_post=1000,
94 | max_num=1000,
95 | nms_thr=0.7,
96 | min_bbox_size=0),
97 | rcnn=dict(
98 | score_thr=0.05, nms=dict(type='nms', iou_thr=0.5), max_per_img=100)
99 | # soft-nms is also supported for rcnn testing
100 | # e.g., nms=dict(type='soft_nms', iou_thr=0.5, min_score=0.05)
101 | )
102 | # dataset settings
103 | dataset_type = 'CocoDataset'
104 | data_root = 'data/coco/'
105 | img_norm_cfg = dict(
106 | mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
107 | data = dict(
108 | imgs_per_gpu=2,
109 | workers_per_gpu=2,
110 | train=dict(
111 | type=dataset_type,
112 | ann_file=data_root + 'annotations/instances_train2017.json',
113 | img_prefix=data_root + 'train2017/',
114 | img_scale=(1333, 800),
115 | img_norm_cfg=img_norm_cfg,
116 | size_divisor=32,
117 | flip_ratio=0.5,
118 | with_mask=False,
119 | with_crowd=True,
120 | with_label=True),
121 | val=dict(
122 | type=dataset_type,
123 | ann_file=data_root + 'annotations/instances_val2017.json',
124 | img_prefix=data_root + 'val2017/',
125 | img_scale=(1333, 800),
126 | img_norm_cfg=img_norm_cfg,
127 | size_divisor=32,
128 | flip_ratio=0,
129 | with_mask=False,
130 | with_crowd=True,
131 | with_label=True),
132 | test=dict(
133 | type=dataset_type,
134 | ann_file=data_root + 'annotations/instances_val2017.json',
135 | img_prefix=data_root + 'val2017/',
136 | img_scale=(1333, 800),
137 | img_norm_cfg=img_norm_cfg,
138 | size_divisor=32,
139 | flip_ratio=0,
140 | with_mask=False,
141 | with_label=False,
142 | test_mode=True))
143 | # optimizer
144 | optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
145 | optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
146 | # learning policy
147 | lr_config = dict(
148 | policy='step',
149 | warmup='linear',
150 | warmup_iters=500,
151 | warmup_ratio=1.0 / 3,
152 | step=[8, 11])
153 | checkpoint_config = dict(interval=1)
154 | # yapf:disable
155 | log_config = dict(
156 | interval=50,
157 | hooks=[
158 | dict(type='TextLoggerHook'),
159 | # dict(type='TensorboardLoggerHook')
160 | ])
161 | # yapf:enable
162 | # runtime settings
163 | total_epochs = 12
164 | dist_params = dict(backend='nccl')
165 | log_level = 'INFO'
166 | work_dir = './work_dirs/faster_rcnn_x101_64x4d_fpn_1x'
167 | load_from = None
168 | resume_from = None
169 | workflow = [('train', 1)]
170 |
--------------------------------------------------------------------------------
/ObjectDetection_TransferLearning_with_mmdetection/utils/make_annot.py:
--------------------------------------------------------------------------------
1 | import os
2 | import pandas as pd
3 | import json
4 | import glob
5 | import numpy as np
6 | import pickle
7 | import cv2
8 | import mmcv
9 | import random
10 | from scipy import io
11 |
12 | def annot(images_path, label_dict, train_data=True):
13 | """Make annotations for mmdetection library from Open Image Dataset Source.
14 | Args:
15 | images_path (str): path to images.
16 | label_dict (dict (str: int)): dictionary with matching class names to class labels.
17 | train_data (boolean): if True - train data, else: validation data.
18 | """
19 | label_path = images_path+'Label/*.txt'
20 | save_val_path = './annotations_val.pkl'
21 | save_train_path = './annotations_train.pkl'
22 | annotations_train = []
23 | annotations_val = []
24 |
25 | # Read labels from .txt files
26 | for name in glob.glob(label_path):
27 | with open(name) as f:
28 | # Create annot dict
29 | annot_instance = {}
30 | # Filename is the path to img
31 | annot_instance['filename'] = images_path+name[:-3].split('/')[-1]+'jpg'
32 | image = cv2.imread(annot_instance['filename'])
33 | h, w = image.shape[:2]
34 | # Height and width of the img
35 | annot_instance['height'] = h
36 | annot_instance['width'] = w
37 | # ann dict inside first dict with boxes and labels for use and boxes and labels for ignore
38 | annot_instance['ann'] = {}
39 | f = [x.rstrip('\n').split() for x in f.readlines()]
40 | kk = [[float(y) for y in x[1:]] for x in f]
41 | ll = [label_dict[x[0]] for x in f]
42 | if len(ll)>0:
43 | annot_instance['ann']['bboxes'] = (np.array(kk)).astype(np.float32)
44 | annot_instance['ann']['labels'] = (np.array(ll)).astype(np.int64)
45 | # If there are no boxes to ignore, this is need to be a zero coordinate vector and empty label vector
46 | annot_instance['ann']['bboxes_ignore'] = np.zeros((0, 4), dtype=np.float32)
47 | annot_instance['ann']['labels_ignore'] = (np.array([])).astype(np.int64)
48 | if train_data:
49 | annotations_train.append(annot_instance)
50 | else:
51 | annotations_val.append(annot_instance)
52 | if train_data:
53 | mmcv.dump(annotations_train, save_train_path)
54 | print('Annot train ready: {}, len {}'.format(save_train_path, len(annotations_train)))
55 | else:
56 | mmcv.dump(annotations_val, save_val_path)
57 | print('Annot val ready: {}, len {}'.format(save_val_path, len(annotations_val)))
58 |
59 | def look_at_imgs_shapes(images_path):
60 | """Collect sizes of images.
61 | """
62 | label_path = images_path+'Label/*.txt'
63 | hh = []
64 | ww = []
65 | for name in glob.glob(label_path):
66 | with open(name) as f:
67 | annot_instance = {}
68 | annot_instance['filename'] = images_path+name[:-3].split('/')[-1]+'jpg'
69 | image = cv2.imread(annot_instance['filename'])
70 | h, w = image.shape[:2]
71 | annot_instance['height'] = h
72 | annot_instance['width'] = w
73 | hh.append(h)
74 | ww.append(w)
75 | return hh, ww
--------------------------------------------------------------------------------
/ObjectDetection_TransferLearning_with_mmdetection/utils/viz.py:
--------------------------------------------------------------------------------
1 | import mmcv
2 | import numpy as np
3 | import cv2
4 | import matplotlib.pyplot as plt
5 | import pycocotools.mask as maskUtils
6 |
7 | def imshow_det_bboxes(img,
8 | bboxes,
9 | labels,
10 | class_names=None,
11 | score_thr=0,
12 | bbox_color='green',
13 | text_color='green',
14 | thickness=2,
15 | font_scale=2,
16 | show=True,
17 | win_name='',
18 | wait_time=0,
19 | out_file=None):
20 | """Draw bboxes and class labels (with scores) on an image.
21 | Args:
22 | img (str or ndarray): The image to be displayed.
23 | bboxes (ndarray): Bounding boxes (with scores), shaped (n, 4) or
24 | (n, 5).
25 | labels (ndarray): Labels of bboxes.
26 | class_names (list[str]): Names of each classes.
27 | score_thr (float): Minimum score of bboxes to be shown.
28 | bbox_color (str or tuple or :obj:`Color`): Color of bbox lines.
29 | text_color (str or tuple or :obj:`Color`): Color of texts.
30 | thickness (int): Thickness of lines.
31 | font_scale (float): Font scales of texts.
32 | show (bool): Whether to show the image.
33 | win_name (str): The window name.
34 | wait_time (int): Value of waitKey param.
35 | out_file (str or None): The filename to write the image.
36 | """
37 | assert bboxes.ndim == 2
38 | assert labels.ndim == 1
39 | assert bboxes.shape[0] == labels.shape[0]
40 | assert bboxes.shape[1] == 4 or bboxes.shape[1] == 5
41 | img = mmcv.imread(img)
42 |
43 | if score_thr > 0:
44 | assert bboxes.shape[1] == 5
45 | scores = bboxes[:, -1]
46 | inds = scores > score_thr
47 | bboxes = bboxes[inds, :]
48 | labels = labels[inds]
49 |
50 | bbox_color = mmcv.color_val(bbox_color)
51 | text_color = mmcv.color_val(text_color)
52 |
53 | for bbox, label in zip(bboxes, labels):
54 | bbox_int = bbox.astype(np.int32)
55 | left_top = (bbox_int[0], bbox_int[1])
56 | right_bottom = (bbox_int[2], bbox_int[3])
57 | cv2.rectangle(
58 | img, left_top, right_bottom, bbox_color, thickness=thickness)
59 | label_text = class_names[
60 | label] if class_names is not None else 'cls {}'.format(label)
61 | if len(bbox) > 4:
62 | label_text += '|{:.02f}'.format(bbox[-1])
63 | cv2.putText(img, label_text, (bbox_int[0], bbox_int[1] - 2),
64 | cv2.FONT_HERSHEY_COMPLEX, font_scale, text_color, thickness)
65 |
66 | if show:
67 | plt.figure(figsize=(20,20))
68 | plt.axis('off')
69 | plt.imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
70 | plt.show()
71 | if out_file is not None:
72 | imwrite(img, out_file)
73 |
74 | def show_result(img,
75 | result,
76 | class_names,
77 | score_thr=0.7,
78 | wait_time=0,
79 | out_file=None,
80 | font_scale=2,
81 | thickness=8,
82 | show_mask=True):
83 | """Visualize the detection results on the image.
84 | Args:
85 | img (str or np.ndarray): Image filename or loaded image.
86 | result (tuple[list] or list): The detection result, can be either
87 | (bbox, segm) or just bbox.
88 | class_names (list[str] or tuple[str]): A list of class names.
89 | score_thr (float): The threshold to visualize the bboxes and masks.
90 | wait_time (int): Value of waitKey param.
91 | out_file (str, optional): If specified, the visualization result will
92 | be written to the out file instead of shown in a window.
93 | """
94 | assert isinstance(class_names, (tuple, list))
95 | img = mmcv.imread(img)
96 | if isinstance(result, tuple):
97 | bbox_result, segm_result = result
98 | else:
99 | bbox_result, segm_result = result, None
100 | bboxes = np.vstack(bbox_result)
101 | # draw segmentation masks
102 | if show_mask and segm_result is not None:
103 | print(len(segm_result[0]))
104 | segms = mmcv.concat_list(segm_result)
105 | inds = np.where(bboxes[:, -1] > score_thr)[0]
106 | for i in inds:
107 | color_mask = np.random.randint(0, 256, (1, 3), dtype=np.uint8)
108 | mask = maskUtils.decode(segms[i]).astype(np.bool)
109 | img[mask] = img[mask] * 0.5 + color_mask * 0.5
110 | # draw bounding boxes
111 | labels = [
112 | np.full(bbox.shape[0], i, dtype=np.int32)
113 | for i, bbox in enumerate(bbox_result)
114 | ]
115 | labels = np.concatenate(labels)
116 | imshow_det_bboxes(
117 | img.copy(),
118 | bboxes,
119 | labels,
120 | class_names=class_names,
121 | score_thr=score_thr,
122 | show=out_file is None,
123 | wait_time=wait_time,
124 | out_file=out_file,
125 | font_scale=font_scale,
126 | thickness=thickness)
--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
1 | # ComputerVision_Tutorials_in_Russian
2 |
3 |
4 | 
5 |
6 |
7 | 1. В папке [ObjectDetection_TransferLearning_with_mmdetection](ObjectDetection_TransferLearning_with_mmdetection/
8 | 'ObjectDetection_TransferLearning_with_mmdetection')
9 | находится туториал по применению
10 | transfer learning для обучения моделей на собственных классах данных с использованием предобученных хребтов
11 | на примере библиотеки [mmdetection](https://github.com/open-mmlab/mmdetection "mmdetection").
12 |
13 | В этом туториале сначала показывается, как можно детектить объекты из классов MS-COCO, например, кота, человека, машину иликухонную утварь.
14 | А потом используются открытые размеченные данные четырех классов, чтобы обучить модель различать выдр, белок, панд и енотов.
15 |
16 |
17 | 
19 | 
21 |
22 |
--------------------------------------------------------------------------------