# Copyright 2021 PaddleFSL 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. from . import CVDataset from paddlefsl.task_sampler import TaskSet import paddlefsl.utils as utils from PIL import Image import numpy as np import random __all__ = ['CifarFS'] class CifarFS(CVDataset): """ Implementation of `CifarFS` dataset. The CifarFS(Cifar100 Few Shot) dataset was introduced by Bertinetto L et al., 2019[1]. CifarFS consists of 100 image classes, each containing 600 samples of size 32*32, RGB format. It was split into 64 classes for training, 16 classes for validating and 20 classes for testing. Users will have to manually confirm and download .zip file and place it properly. Download URL: https://drive.google.com/file/d/1nN1u2ZeD0L90uG5Y_Ml6uQR6z-o6aBLL/view Refs: 1. Bertinetto L et al. 2019. “Meta-learning with differentiable closed-form solvers” ICLR. Args: mode(str): mode of the datasets, whether used as meta-training, meta-validation or meta-testing. image_size(tuple, optional): size of the single image, for example (28, 28), default None. If None, this code will return the original image size. root(str, optional): root directory of the .zip file, can be set None. If None, it will be set default root: '/paddlefsl/../raw_data/'. This code will check whether root contains .zip file. If not, error occurs to inform the user to download file manually. transform(callable, optional): transform to perform on image, None for no transform. backend(str, optional): Specifies which type of image to be returned: PIL.Image or numpy.ndarray. Should be one of {'pil', 'numpy'}. Default 'numpy'. Returns: Dataset: instance of CifarFS datasets Examples: ..code-block:: python from paddlefsl.vision.datasets import CifarFS training_set = CifarFS(mode='train') validating_set = CifarFS(mode='valid', backend='pil') image, label = training_set[16] print(image.shape) # (3, 32, 32) print(label) # 'train' image, label = validating_set[15] image.show() # shows the image of an otter print(label) # 'otter' print(len(training_set)) # 64 * 600 = 38400 task = training_set.sample_task_set(ways=5, shots=5) print(task.support_images.shape) # (25, 3, 32, 32) """ zip_url = 'https://drive.google.com/file/d/1nN1u2ZeD0L90uG5Y_Ml6uQR6z-o6aBLL/view' zip_name = 'cifar100.zip' zip_md5 = 'fdb4405027b809aa4c059047f21ca3d1' def __init__(self, mode, root=None, image_size=None, transform=None, backend='numpy'): super(CifarFS, self).__init__('cifar100', mode, root, image_size, transform, backend) self._check_files() # self._label_image_names: List[ (label_name, List[image names]) ] self._labels_image_names = self._load_data() def __getitem__(self, idx): label_idx, image_idx = int(idx / 600), idx % 600 label_name = self._labels_image_names[label_idx][0] image_name = self._labels_image_names[label_idx][1][image_idx] image = self._get_image_by_name(label_name, image_name) return image, label_name def __len__(self): return len(self._labels_image_names) * 600 def sample_task_set(self, ways=5, shots=5, query_num=None): query_num = shots if query_num is None else query_num sample = random.sample(self._labels_image_names, ways) # result: List[ (str(label name), List[np.ndarray(image)]) ] result = [] for (label_name, image_names) in sample: image_list = [self._get_image_by_name(label_name, image_name, backend='numpy') for image_name in random.sample(image_names, shots + query_num)] result.append((label_name, image_list)) return TaskSet(label_names_data=result, ways=ways, shots=shots, query_num=query_num) def _check_files(self): if utils.check_file_exist(self.root + '/..', self.zip_name, self.zip_md5): print("Using downloaded and verified .zip file. Decompressing...") utils.decompress(self.root + '/../' + self.zip_name, self.root + '/..') print("Decompress finished.") else: error_info = "Data files not found. Please download CIFAR-FS file manually from\n" \ + self.zip_url + "\nand place the .zip file under 'root' which you can set, " \ + "or under default root: " + utils.DATA_HOME raise RuntimeError(error_info) def _load_data(self): mode = 'val' if self.mode == 'valid' else self.mode with open(self.root + '/splits/bertinetto/' + mode + '.txt') as f: label_names = f.read().split('\n')[:-1] label_image_names = [(label, [image for image in utils.list_files(self.root + '/data/' + label, '.png')]) for label in label_names] return label_image_names def _get_image_by_name(self, label_name, image_name, backend=None): backend = self.backend if backend is None else backend image = Image.open(self.root + '/data/' + label_name + '/' + image_name, mode='r').convert('RGB') if self.image_size is not None: image = image.resize(self.image_size, Image.ANTIALIAS) if backend == 'numpy': image = np.array(image) image = np.transpose(image, [2, 0, 1]) if self.transform is not None: image = self.transform(image) return image