def create_many_feature_map_files_from_config(config, split_name):
"""
Reads the image files from the sub-directories given as split names.
Creates the according feature map files in the top directory.
"""
run_opts = config.run_opts
print("Run opts: " + str(run_opts))
batch_size = __get_batch_size(config, run_opts)
expected_num_images = __get_expected_num_images(config, run_opts,
split_name)
is_dryrun = __is_dryrun(run_opts)
images_top_directory = config.getDatasetImagesDirectoryPath()
target_shape = config.getImageInputShape()
image_features_size = config.getImageFeaturesSize()
image_infix = get_infix_from_config(config, split_name)
image_prefix = "COCO_" + image_infix
output_layer = config.getImageOutputLayer()
return create_many_feature_map_files(images_top_directory, output_layer,
image_prefix, image_features_size,
target_shape, batch_size, split_name,
expected_num_images, is_dryrun)
def visualize_images_with_caption_by_image_ids(config, sample_image_ids, sample_results, image_rows, image_cols, split_name):
directory_path = "/".join([config.getDatasetImagesDirectoryPath(), split_name])
image_infix = get_infix_from_config(config, split_name)
captions = to_caption_listing_by_image_id(load_prepared_captions_json_from_config(config, split_name))
provider = MscocoFileSystemImageProvider(directory_path, prefix="COCO_" + image_infix, vgg_like=False, target_size=(448, 448))
sample_images = provider.get_images_for_image_ids(sample_image_ids)
sample_images = sample_images.astype('uint8')
titles = ["{:d}: {}".format(image_id, captions[image_id][0]) for image_id in sample_image_ids]
titles = ["\n".join([titles[idx], r]) for idx, r in enumerate(sample_results)]
show_many(sample_images, image_rows, image_cols, figsize=(4, 3), titles=titles)
def create_single_split_provider_from_config(config, split_name):
directory_path = "/".join(
[config.getDatasetImagesDirectoryPath(), split_name])
image_infix = get_infix_from_config(config, split_name)
if config.getByPassImageFeaturesComputation():
return MscocoFileSystemFeatureMapProvider(directory_path,
prefix="COCO_" +
image_infix)
else:
return MscocoFileSystemImageProvider(
directory_path,
prefix="COCO_" + image_infix,
vgg_like=True,
target_size=config.getImageInputShape())
def create_from_config(config, split_name):
directory_path = "/".join([config.getDatasetImagesDirectoryPath(), split_name])
image_prefix = "COCO_" + get_infix_from_config(config, split_name)
print("Image prefix:", image_prefix)
caption_max_length = config.getCaptionMaximalLength()
print("Caption max length:", caption_max_length)
by_pass = config.getByPassImageFeaturesComputation()
print("Bypass image feature compuration:", by_pass)
target_shape = config.getImageInputShape()
print("Target image shape:", target_shape)
feature_size = config.getImageFeaturesSize()
print("Image feature size:", feature_size)
return ExternAttentionExperimentSequence(directory_path, image_prefix, caption_max_length, by_pass, target_shape, feature_size)
def create_many_attention_map_files_from_config(config, split_name):
"""
Reads the image files from the sub-directories given as split names.
Creates the according feature map files in the top directory.
"""
target_shape = config.getImageInputShape()
image_feature_size = config.getImageFeaturesSize()
image_infix = get_infix_from_config(config, split_name)
image_prefix = "COCO_" + image_infix
bounding_boxes = load_prepared_boxes_json_from_config(config, split_name)
boxes = calculate_metrics(bounding_boxes)
boxes_by_id = collections.defaultdict(list)
[boxes_by_id[box["image_id"]].append(box) for box in boxes]
images_top_directory = config.getDatasetImagesDirectoryPath()
image_paths = get_image_paths(to_split_dir(images_top_directory, split_name))
processables = to_processables(image_paths, boxes_by_id, target_shape, image_prefix, image_feature_size)
preprocess_bounding_boxes(processables)