(img, bbox, data_split,data,dataname)
| 198 | return img, trans, inv_trans, rot, do_flip |
| 199 | |
| 200 | def augmentation_instance_sample(img, bbox, data_split,data,dataname): |
| 201 | ori_shape = img.shape[:2][::-1] |
| 202 | |
| 203 | if getattr(cfg, 'no_aug', False) and data_split == 'train': |
| 204 | scale, rot, color_scale, do_flip,size,crop,sample_ratio,sample_prob = 1.0, 0.0, np.array([1, 1, 1]), False, ori_shape, np.array([1,1]), 0,0 |
| 205 | |
| 206 | size = random.choice(cfg.train_sizes) |
| 207 | max_size = cfg.train_max_size |
| 208 | elif data_split == 'train': |
| 209 | scale, rot, color_scale, do_flip, crop, sample_ratio,sample_prob = get_aug_config(dataname) |
| 210 | rot=0 |
| 211 | # scale, rot, do_flip, crop = 1.0, 0.0, False, np.array([1,1]) |
| 212 | size = random.choice(cfg.train_sizes) |
| 213 | max_size = cfg.train_max_size |
| 214 | else: |
| 215 | scale, rot, color_scale, do_flip, crop,sample_ratio,sample_prob = 1.0, 0.0, np.array([1, 1, 1]), False, np.array([1,1]),0,0 |
| 216 | size = random.choice(cfg.test_sizes) |
| 217 | max_size = cfg.test_max_size |
| 218 | |
| 219 | |
| 220 | if random.random() < sample_prob: |
| 221 | crop_person_number = len(data['bbox']) |
| 222 | |
| 223 | if random.random() < sample_ratio: |
| 224 | if random.random() < 0.6: |
| 225 | crop_person_number_sample = 1 |
| 226 | else: |
| 227 | crop_person_number_sample = np.random.randint(crop_person_number) + 1 |
| 228 | else: |
| 229 | crop_person_number_sample = crop_person_number |
| 230 | sample_ids = np.array( |
| 231 | random.sample(list(range(crop_person_number)), crop_person_number_sample)) |
| 232 | |
| 233 | bbox_xyxy = [] |
| 234 | |
| 235 | bbox_xyxy = np.stack(data['bbox'],axis=0)[sample_ids] |
| 236 | |
| 237 | leftTop_ = bbox_xyxy[:, :2] |
| 238 | leftTop_ = np.array([np.min(leftTop_[:, 0]), np.min(leftTop_[:, 1])]) |
| 239 | rightBottom_ = bbox_xyxy[:, 2:4] |
| 240 | rightBottom_ = np.array( |
| 241 | [np.max(rightBottom_[:, 0]), |
| 242 | np.max(rightBottom_[:, 1])]) |
| 243 | crop_bbox_xyxy = np.concatenate([leftTop_, rightBottom_]) |
| 244 | crop_bbox_xywh = crop_bbox_xyxy.copy() |
| 245 | crop_bbox_xywh[2:] = crop_bbox_xywh[2:]-crop_bbox_xywh[:2] |
| 246 | crop_bbox_xywh = adjust_bounding_box(crop_bbox_xywh,ori_shape[0],ori_shape[1]) |
| 247 | else: |
| 248 | crop_bbox_xywh = bbox.copy() |
| 249 | reshape_size = resize(crop_bbox_xywh[2:], size, max_size) |
| 250 | # try: |
| 251 | # reshape_size = resize(crop_bbox_xywh[2:], size, max_size) |
| 252 | # except Exception as e: |
| 253 | # print(crop_bbox_xywh) |
| 254 | # print(size) |
| 255 | # print(max_size) |
| 256 | # raise e |
| 257 | img, trans, inv_trans = generate_patch_image(img, crop_bbox_xywh, 1, rot, do_flip, reshape_size[::-1]) |
no test coverage detected