T2IDataset.py

#encoding=utf8
# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.

# 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.
import os
import io
from paddle.io import Dataset, DataLoader
import h5py
import numpy as np
import pdb
from PIL import Image
import paddle
from paddle.vision import transforms


# load the dataset
class Text2ImageDataset(Dataset):
    def __init__(self, datasetFile, transform=None, split=0):
        # initialize the dataset
        super().__init__()
        self.datasetFile = datasetFile
        self.transform = transforms.Compose([
            transforms.ToTensor(),
            transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
        ])
        self.dataset = None
        self.dataset_keys = None
        self.split = 'train' if split == 0 else 'valid' if split == 1 else 'test'
        self.h5py2int = lambda x: int(np.array(x))

    def __len__(self):
        # get the total number of data
        f = h5py.File(self.datasetFile, 'r')
        self.dataset_keys = [str(k) for k in f[self.split].keys()]
        length = len(f[self.split])
        f.close()

        return length

    def __getitem__(self, idx):
        """
        :return：
        a dictionary of data
        sample = {
            'right_images': real_image,
            'right_embed': sentence_embeddings,
            'wrong_images': wrong_image,
            'inter_embed': interpolated_embeddings,
            'txt': sentences
        }
        """
        if self.dataset is None:
            self.dataset = h5py.File(self.datasetFile, mode='r')
            self.dataset_keys = [str(k) for k in self.dataset[self.split].keys()]

        example_name = self.dataset_keys[idx]
        example = self.dataset[self.split][example_name]
        right_image = bytes(np.array(example['img']))
        right_embed = np.array(example['embeddings'], dtype=float)
        wrong_image = bytes(np.array(self.find_wrong_image(example['class'])))
        inter_embed = np.array(self.find_inter_embed())

        right_image = Image.open(io.BytesIO(right_image)).resize((64, 64))
        wrong_image = Image.open(io.BytesIO(wrong_image)).resize((64, 64))

        txt = np.array(example['txt']).astype(str)

        right_image = self.transform(right_image)
        wrong_image = self.transform(wrong_image)

        sample = {
            'right_images': right_image,
            'right_embed': paddle.to_tensor(right_embed, dtype='float32'),
            'wrong_images': wrong_image,
            'inter_embed': paddle.to_tensor(inter_embed, dtype='float32'),
            'txt': str(txt)
        }

        return sample

    def find_wrong_image(self, category):
        # randomly select the wrong images for sentence embeddings
        idx = np.random.randint(len(self.dataset_keys))
        example_name = self.dataset_keys[idx]
        example = self.dataset[self.split][example_name]
        _category = example['class']

        if _category != category:
            return example['img']

        return self.find_wrong_image(category)

    def find_inter_embed(self):
        # get the interpolated sentence embeddings
        idx = np.random.randint(len(self.dataset_keys))
        example_name = self.dataset_keys[idx]
        example = self.dataset[self.split][example_name]
        return example['embeddings']