# coding=utf-8 # Copyright 2020 The TensorFlow Datasets Authors. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # Lint as: python3 """Visual Domain Decathlon Datasets.""" from __future__ import absolute_import from __future__ import division from __future__ import print_function import json import os import tensorflow.compat.v2 as tf import tensorflow_datasets.public_api as tfds _DESCRIPTION = """\ This contains the 10 datasets used in the Visual Domain Decathlon, part of the PASCAL in Detail Workshop Challenge (CVPR 2017). The goal of this challenge is to solve simultaneously ten image classification problems representative of very different visual domains. Some of the datasets included here are also available as separate datasets in TFDS. However, notice that images were preprocessed for the Visual Domain Decathlon (resized isotropically to have a shorter size of 72 pixels) and might have different train/validation/test splits. Here we use the official splits for the competition. """ _CITATION = """\ @ONLINE{hakanbilensylvestrerebuffitomasjakab2017, author = "Hakan Bilen, Sylvestre Rebuffi, Tomas Jakab", title = "Visual Domain Decathlon", year = "2017", url = "https://www.robots.ox.ac.uk/~vgg/decathlon/" } """ _URL_PREFIX_VGG = 'http://www.robots.ox.ac.uk/~vgg/share/' _URL_PREFIX_IMAGENET = 'http://www.image-net.org/image/decathlon/' _CONFIG_DESCRIPTION_PATTERN = ( 'Data based on "{}", with images resized isotropically to have a shorter ' 'size of 72 pixels.') class VisualDomainDecathlonConfig(tfds.core.BuilderConfig): def __init__(self, num_classes, **kwargs): self.num_classes = num_classes if 'version' not in kwargs: kwargs['version'] = tfds.core.Version('1.2.0') super(VisualDomainDecathlonConfig, self).__init__(**kwargs) def _get_builder_configs(): """Returns the list of builder configs for the dataset.""" configs = [] for short_name, full_name, num_classes in [ ('aircraft', 'Aircraft', 100), ('cifar100', 'CIFAR-100', 100), ('daimlerpedcls', 'Daimler Pedestrian Classification', 2), ('dtd', 'Describable Textures', 47), ('gtsrb', 'German Traffic Signs', 43), ('imagenet12', 'Imagenet', 1000), ('omniglot', 'Omniglot', 1623), ('svhn', 'Street View House Numbers', 10), ('ucf101', 'UCF101 Dynamic Images', 101), ('vgg-flowers', 'VGG-Flowers', 102), ]: description = _CONFIG_DESCRIPTION_PATTERN.format(full_name) configs.append(VisualDomainDecathlonConfig(name=short_name, num_classes=num_classes, description=description)) return configs class VisualDomainDecathlon(tfds.core.GeneratorBasedBuilder): """Visual Domain Decathlon Datasets.""" BUILDER_CONFIGS = _get_builder_configs() def _info(self): num_classes = self.builder_config.num_classes return tfds.core.DatasetInfo( builder=self, description=_DESCRIPTION, features=tfds.features.FeaturesDict({ 'name': tfds.features.Text(), 'image': tfds.features.Image(shape=(None, None, 3), encoding_format='jpeg'), 'label': tfds.features.ClassLabel(num_classes=num_classes), }), supervised_keys=('image', 'label'), homepage='https://www.robots.ox.ac.uk/~vgg/decathlon/', citation=_CITATION, ) def _split_generators(self, dl_manager): if self.builder_config.name == 'imagenet12': devkit_path, images_archive = dl_manager.download_and_extract([ _URL_PREFIX_VGG + 'decathlon-1.0-devkit.tar.gz', tfds.download.Resource( url=_URL_PREFIX_IMAGENET + 'decathlon-1.0-data-imagenet.tar', extract_method=tfds.download.ExtractMethod.NO_EXTRACT), ]) else: devkit_path, data_path = dl_manager.download_and_extract([ _URL_PREFIX_VGG + 'decathlon-1.0-devkit.tar.gz', _URL_PREFIX_VGG + 'decathlon-1.0-data.tar.gz', ]) images_archive = os.path.join(data_path, self.builder_config.name + '.tar') annotations_path = os.path.join(devkit_path, 'decathlon-1.0', 'annotations') return [ tfds.core.SplitGenerator( name=tfds.Split.TRAIN, gen_kwargs=dict( images_archive=images_archive, annotations_path=annotations_path, split='train'), ), tfds.core.SplitGenerator( name=tfds.Split.TEST, gen_kwargs=dict( images_archive=images_archive, annotations_path=annotations_path, split='test'), ), tfds.core.SplitGenerator( name=tfds.Split.VALIDATION, gen_kwargs=dict( images_archive=images_archive, annotations_path=annotations_path, split='val'), ), ] def _generate_examples(self, images_archive, annotations_path, split): """Yields examples.""" filename_to_label = _get_filename_to_label_map( annotations_path=annotations_path, dataset_name=self.builder_config.name, split=split) for image_fname, image_fobj in tfds.download.iter_archive( path=images_archive, method=tfds.download.ExtractMethod.TAR_STREAM): image_fname = image_fname.replace('\\', '/') # For windows compatibility if image_fname in filename_to_label: label = filename_to_label[image_fname] example = { 'name': image_fname, 'image': image_fobj, 'label': label, } yield image_fname, example def _get_filename_to_label_map(annotations_path, dataset_name, split): """Returns a mapping from image filenames to labels, for the given split.""" filename_to_label = {} if split == 'test': filepath = os.path.join(annotations_path, dataset_name + '_test_stripped.json') else: filepath = os.path.join(annotations_path, dataset_name + '_%s.json' % split) prefix = 'data/' with tf.io.gfile.GFile(filepath, mode='r') as f: annotations = json.load(f) if split == 'test': # For test, labels are unknown. for example_info in annotations['images']: image_filename = example_info['file_name'] image_filename = image_filename[len(prefix):] filename_to_label[image_filename] = -1 else: # Load a map from category ID to label index. category_id_to_label = {} for i, category_info in enumerate(annotations['categories']): category_id_to_label[category_info['id']] = i # Load a map from image ID to image filename. image_id_to_filename = {} for example_info in annotations['images']: image_id_to_filename[example_info['id']] = example_info['file_name'] # Load the map from image filename to label. for example_info in annotations['annotations']: image_filename = image_id_to_filename[example_info['image_id']] image_filename = image_filename[len(prefix):] label = category_id_to_label[example_info['category_id']] filename_to_label[image_filename] = label return filename_to_label