def test_random_resized_crop_with_bbox_op_bad_c(): """ Test RandomCropWithBBox op with invalid bounding boxes, expected to catch multiple errors. """ logger.info("test_random_resized_crop_with_bbox_op_bad_c") test_op = c_vision.RandomResizedCropWithBBox((256, 512), (0.5, 0.5), (0.5, 0.5)) data_voc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False) check_bad_bbox(data_voc2, test_op, InvalidBBoxType.WidthOverflow, "bounding boxes is out of bounds of the image") data_voc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False) check_bad_bbox(data_voc2, test_op, InvalidBBoxType.HeightOverflow, "bounding boxes is out of bounds of the image") data_voc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False) check_bad_bbox(data_voc2, test_op, InvalidBBoxType.NegativeXY, "min_x") data_voc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False) check_bad_bbox(data_voc2, test_op, InvalidBBoxType.WrongShape, "4 features")
def test_c_random_resized_crop_with_bbox_op_invalid2(): """ Prints images side by side with and without Aug applied + bboxes to compare and test """ # Load dataset # only loading the to AugDataset as test will fail on this dataVoc2 = ds.VOCDataset(DATA_DIR, task="Detection", mode="train", decode=True, shuffle=False) try: # If input range of ratio is not in the order of (min, max), ValueError will be raised. test_op = c_vision.RandomResizedCropWithBBox((256, 512), (1, 1), (1, 0.5)) # maps to fix annotations to HQ standard dataVoc2 = dataVoc2.map(input_columns=["annotation"], output_columns=["annotation"], operations=fix_annotate) # map to apply ops dataVoc2 = dataVoc2.map(input_columns=["image", "annotation"], output_columns=["image", "annotation"], columns_order=["image", "annotation"], operations=[test_op]) for _ in dataVoc2.create_dict_iterator(): break except ValueError as err: logger.info("Got an exception in DE: {}".format(str(err))) assert "Input range is not valid" in str(err)
def test_random_resized_crop_with_bbox_op_invalid2_c(): """ Tests RandomResizedCropWithBBox Op on invalid constructor parameters, expected to raise ValueError """ logger.info("test_random_resized_crop_with_bbox_op_invalid2_c") # Load dataset # only loading the to AugDataset as test will fail on this dataVoc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False) try: # If input range of ratio is not in the order of (min, max), ValueError will be raised. test_op = c_vision.RandomResizedCropWithBBox((256, 512), (1, 1), (1, 0.5)) # map to apply ops dataVoc2 = dataVoc2.map(input_columns=["image", "annotation"], output_columns=["image", "annotation"], columns_order=["image", "annotation"], operations=[test_op]) for _ in dataVoc2.create_dict_iterator(): break except ValueError as err: logger.info("Got an exception in DE: {}".format(str(err))) assert "Input is not within the required interval of (0 to 16777216)." in str( err)
def test_random_resized_crop_with_bbox_op_edge_c(plot_vis=False): """ Prints images and bboxes side by side with and without RandomResizedCropWithBBox Op applied, tests on dynamically generated edge case, expected to pass """ logger.info("test_random_resized_crop_with_bbox_op_edge_c") # Load dataset dataVoc1 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False) dataVoc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False) test_op = c_vision.RandomResizedCropWithBBox((256, 512), (0.5, 0.5), (0.5, 0.5)) # maps to convert data into valid edge case data dataVoc1 = dataVoc1.map( input_columns=["image", "annotation"], output_columns=["image", "annotation"], columns_order=["image", "annotation"], operations=[ lambda img, bboxes: (img, np.array([[0, 0, img.shape[1], img.shape[0]]]).astype(bboxes. dtype)) ]) # Test Op added to list of Operations here dataVoc2 = dataVoc2.map( input_columns=["image", "annotation"], output_columns=["image", "annotation"], columns_order=["image", "annotation"], operations=[ lambda img, bboxes: (img, np.array([[0, 0, img.shape[1], img.shape[0]]]).astype( bboxes.dtype)), test_op ]) unaugSamp, augSamp = [], [] for unAug, Aug in zip(dataVoc1.create_dict_iterator(), dataVoc2.create_dict_iterator()): unaugSamp.append(unAug) augSamp.append(Aug) if plot_vis: visualize_with_bounding_boxes(unaugSamp, augSamp)
def test_c_random_resized_crop_with_bbox_op_edge(plot_vis=False): """ Prints images side by side with and without Aug applied + bboxes to compare and test """ # Load dataset dataVoc1 = ds.VOCDataset(DATA_DIR, task="Detection", mode="train", decode=True, shuffle=False) dataVoc2 = ds.VOCDataset(DATA_DIR, task="Detection", mode="train", decode=True, shuffle=False) test_op = c_vision.RandomResizedCropWithBBox((256, 512), (0.5, 0.5), (0.5, 0.5)) # maps to fix annotations to HQ standard dataVoc1 = dataVoc1.map(input_columns=["annotation"], output_columns=["annotation"], operations=fix_annotate) dataVoc2 = dataVoc2.map(input_columns=["annotation"], output_columns=["annotation"], operations=fix_annotate) # Modify BBoxes to serve as valid edge cases dataVoc2 = dataVoc2.map(input_columns=["image", "annotation"], output_columns=["image", "annotation"], columns_order=["image", "annotation"], operations=[gen_bbox_edge]) # map to apply ops dataVoc2 = dataVoc2.map(input_columns=["image", "annotation"], output_columns=["image", "annotation"], columns_order=["image", "annotation"], operations=[test_op ]) # Add column for "annotation" unaugSamp, augSamp = [], [] for unAug, Aug in zip(dataVoc1.create_dict_iterator(), dataVoc2.create_dict_iterator()): unaugSamp.append(unAug) augSamp.append(Aug) if plot_vis: visualize(unaugSamp, augSamp)
def test_random_resized_crop_with_bbox_op_c(plot_vis=False): """ Prints images and bboxes side by side with and without RandomResizedCropWithBBox Op applied, tests with MD5 check, expected to pass """ logger.info("test_random_resized_crop_with_bbox_op_c") original_seed = config_get_set_seed(23415) original_num_parallel_workers = config_get_set_num_parallel_workers(1) # Load dataset dataVoc1 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False) dataVoc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False) test_op = c_vision.RandomResizedCropWithBBox((256, 512), (0.5, 0.5), (0.5, 0.5)) # map to apply ops dataVoc2 = dataVoc2.map(input_columns=["image", "annotation"], output_columns=["image", "annotation"], columns_order=["image", "annotation"], operations=[test_op]) filename = "random_resized_crop_with_bbox_01_c_result.npz" save_and_check_md5(dataVoc2, filename, generate_golden=GENERATE_GOLDEN) unaugSamp, augSamp = [], [] for unAug, Aug in zip(dataVoc1.create_dict_iterator(), dataVoc2.create_dict_iterator()): unaugSamp.append(unAug) augSamp.append(Aug) if plot_vis: visualize_with_bounding_boxes(unaugSamp, augSamp) # Restore config setting ds.config.set_seed(original_seed) ds.config.set_num_parallel_workers(original_num_parallel_workers)
def test_random_resized_crop_with_bbox_op_coco_c(plot_vis=False): """ Prints images and bboxes side by side with and without RandomResizedCropWithBBox Op applied, Testing with Coco dataset """ logger.info("test_random_resized_crop_with_bbox_op_coco_c") # load dataset dataCoco1 = ds.CocoDataset(DATA_DIR_COCO[0], annotation_file=DATA_DIR_COCO[1], task="Detection", decode=True, shuffle=False) dataCoco2 = ds.CocoDataset(DATA_DIR_COCO[0], annotation_file=DATA_DIR_COCO[1], task="Detection", decode=True, shuffle=False) test_op = c_vision.RandomResizedCropWithBBox((512, 512), (0.5, 1), (0.5, 1)) dataCoco2 = dataCoco2.map(input_columns=["image", "bbox"], output_columns=["image", "bbox"], columns_order=["image", "bbox"], operations=[test_op]) unaugSamp, augSamp = [], [] for unAug, Aug in zip(dataCoco1.create_dict_iterator(), dataCoco2.create_dict_iterator()): unaugSamp.append(unAug) augSamp.append(Aug) if plot_vis: visualize_with_bounding_boxes(unaugSamp, augSamp, "bbox")