(img, bbox=None, input_resolution=(256, 192), rescale=1.25, mask=None, **kwargs)
| 313 | |
| 314 | @staticmethod |
| 315 | def preprocess(img, bbox=None, input_resolution=(256, 192), rescale=1.25, mask=None, **kwargs): |
| 316 | if bbox is None or bbox[-1] <= 0 or (bbox[2] - bbox[0]) < 10 or (bbox[3] - bbox[1]) < 10: |
| 317 | bbox = np.array([0, 0, img.shape[1], img.shape[0]]) |
| 318 | |
| 319 | bbox_xywh = bbox |
| 320 | if mask is not None: |
| 321 | img = np.where(mask > 128, img, mask) |
| 322 | |
| 323 | if isinstance(input_resolution, int): |
| 324 | center, scale = bbox_from_detector(bbox_xywh, (input_resolution, input_resolution), rescale=rescale) |
| 325 | img, new_shape, old_xy, new_xy = crop(img, center, scale, (input_resolution, input_resolution)) |
| 326 | else: |
| 327 | center, scale = bbox_from_detector(bbox_xywh, input_resolution, rescale=rescale) |
| 328 | img, new_shape, old_xy, new_xy = crop(img, center, scale, (input_resolution[0], input_resolution[1])) |
| 329 | |
| 330 | IMG_NORM_MEAN = np.array([0.485, 0.456, 0.406]) |
| 331 | IMG_NORM_STD = np.array([0.229, 0.224, 0.225]) |
| 332 | img_norm = (img / 255.0 - IMG_NORM_MEAN) / IMG_NORM_STD |
| 333 | img_norm = img_norm.transpose(2, 0, 1).astype(np.float32) |
| 334 | return img_norm, np.array(center), np.array(scale) |
| 335 | |
| 336 | |
| 337 | class Pose2d: |
nothing calls this directly
no test coverage detected