| 329 | |
| 330 | |
| 331 | def test_coco_empty_annotations_pix(): |
| 332 | file_root = os.path.join(test_data_root, "db", "coco_dummy", "images") |
| 333 | train_annotations = os.path.join(test_data_root, "db", "coco_dummy", "instances.json") |
| 334 | |
| 335 | @pipeline_def(batch_size=1, device_id=0, num_threads=4) |
| 336 | def coco_pipe(): |
| 337 | _, _, _, masks, ids = fn.readers.coco( |
| 338 | file_root=file_root, |
| 339 | annotations_file=train_annotations, |
| 340 | image_ids=True, |
| 341 | pixelwise_masks=True, |
| 342 | ) |
| 343 | return masks, ids |
| 344 | |
| 345 | pipe = coco_pipe() |
| 346 | number_of_samples = pipe.epoch_size() |
| 347 | for k in number_of_samples: |
| 348 | # there is only one reader |
| 349 | number_of_samples = number_of_samples[k] |
| 350 | break |
| 351 | |
| 352 | annotations = None |
| 353 | with open(train_annotations) as file: |
| 354 | annotations = json.load(file) |
| 355 | |
| 356 | anno_mapping = {} |
| 357 | for elm in annotations["annotations"]: |
| 358 | image_id = elm["image_id"] |
| 359 | anno_mapping[image_id] = anno_mapping.get(image_id, False) or "segmentation" in elm |
| 360 | |
| 361 | for _ in range(number_of_samples): |
| 362 | mask, image_ids = pipe.run() |
| 363 | image_ids = int(image_ids.as_array().item()) |
| 364 | max_mask = np.max(np.array(mask.as_tensor())) |
| 365 | assert (max_mask != 0 and image_ids in anno_mapping and anno_mapping[image_ids]) or ( |
| 366 | max_mask == 0 and not (image_ids in anno_mapping and anno_mapping[image_ids]) |
| 367 | ) |
| 368 | |
| 369 | |
| 370 | def test_coco_empty_annotations_poly(): |