mlebench/competitions/herbarium-2020-fgvc7/prepare.py [152:184]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    new_test_metadata.update(
        {
            "annotations": [],
            "images": [],
        }
    )
    # Flatten and shuffle test set so that we don't have all the same categories in a row
    test_annotations_images = [
        item for sublist in test_annotations_images_by_category.values() for item in sublist
    ]
    random.Random(0).shuffle(test_annotations_images)
    for idx, annotation_image in tqdm(
        enumerate(test_annotations_images),
        desc="Creating new test dataset",
        total=len(test_annotations_images),
    ):

        # Make new image id, for test set this is just the index
        new_image_id = str(idx)
        # Make new filename from image id e.g. "000/0.jpg"
        new_file_name = f"images/{idx // 1000:03d}/{idx}.jpg"

        new_annotation = annotation_image["annotation"].copy()
        new_annotation["image_id"] = new_image_id
        new_test_metadata["annotations"].append(new_annotation)

        new_image = annotation_image["image"].copy()
        new_image["id"] = new_image_id
        new_image["file_name"] = new_file_name
        new_test_metadata["images"].append(new_image)

        # Copy file from raw to public
        if not dev_mode or idx < dev_count:  # if dev_mode, only copy the first dev_count images
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



mlebench/competitions/herbarium-2021-fgvc8/prepare.py [152:184]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    new_test_metadata.update(
        {
            "annotations": [],
            "images": [],
        }
    )
    # Flatten and shuffle test set so that we don't have all the same categories in a row
    test_annotations_images = [
        item for sublist in test_annotations_images_by_category.values() for item in sublist
    ]
    random.Random(0).shuffle(test_annotations_images)
    for idx, annotation_image in tqdm(
        enumerate(test_annotations_images),
        desc="Creating new test dataset",
        total=len(test_annotations_images),
    ):

        # Make new image id, for test set this is just the index
        new_image_id = str(idx)
        # Make new filename from image id e.g. "000/0.jpg"
        new_file_name = f"images/{idx // 1000:03d}/{idx}.jpg"

        new_annotation = annotation_image["annotation"].copy()
        new_annotation["image_id"] = new_image_id
        new_test_metadata["annotations"].append(new_annotation)

        new_image = annotation_image["image"].copy()
        new_image["id"] = new_image_id
        new_image["file_name"] = new_file_name
        new_test_metadata["images"].append(new_image)

        # Copy file from raw to public
        if not dev_mode or idx < dev_count:  # if dev_mode, only copy the first dev_count images
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -