dataset.py

import os.path as osp
import random

import cv2
import numpy as np
import paddle
from paddle.io import Dataset

from utils.misc import scandir
from utils.file_client import FileClient
from utils.img_utils import padding, img2tensor, imfrombytes
from data.transforms import paired_random_crop, random_augmentation

class Dataset_GaussianDenoising(Dataset):
    """Paired image dataset for image restoration.

    Read LQ (Low Quality, e.g. LR (Low Resolution), blurry, noisy, etc) and
    GT image pairs.

    There are three modes:
    1. 'lmdb': Use lmdb files.
        If opt['io_backend'] == lmdb.
    2. 'meta_info_file': Use meta information file to generate paths.
        If opt['io_backend'] != lmdb and opt['meta_info_file'] is not None.
    3. 'folder': Scan folders to generate paths.
        The rest.

    Args:
        opt (dict): Config for train datasets. It contains the following keys:
            dataroot_gt (str): Data root path for gt.
            meta_info_file (str): Path for meta information file.
            io_backend (dict): IO backend type and other kwarg.
            gt_size (int): Cropped patched size for gt patches.
            use_flip (bool): Use horizontal flips.
            use_rot (bool): Use rotation (use vertical flip and transposing h
                and w for implementation).

            scale (bool): Scale, which will be added automatically.
            phase (str): 'train' or 'val'.
    """

    def __init__(self, opt):
        super(Dataset_GaussianDenoising, self).__init__()
        self.opt = opt

        if self.opt['phase'] == 'train':
            self.sigma_type  = opt['sigma_type']
            self.sigma_range = opt['sigma_range']
            assert self.sigma_type in ['constant', 'random', 'choice']
        else:
            self.sigma_test = opt['sigma_test']
        self.in_ch = opt['in_ch']

        # file client (io backend)
        self.file_client = None
        self.io_backend_opt = opt['io_backend']
        self.mean = opt['mean'] if 'mean' in opt else None
        self.std = opt['std'] if 'std' in opt else None

        self.gt_folder = opt['dataroot_gt']

        self.paths = sorted(list(scandir(self.gt_folder, full_path=True)))

        if self.opt['phase'] == 'train':
            self.geometric_augs = self.opt['geometric_augs']

    def __getitem__(self, index):
        if self.file_client is None:
            self.file_client = FileClient(
                self.io_backend_opt.pop('type'), **self.io_backend_opt)

        scale = self.opt['scale']
        index = index % len(self.paths)
        # Load gt and lq images. Dimension order: HWC; channel order: BGR;
        # image range: [0, 1], float32.
        gt_path = self.paths[index]
        img_bytes = self.file_client.get(gt_path, 'gt')

        if self.in_ch == 3:
            try:
                img_gt = imfrombytes(img_bytes, float32=True)
            except:
                raise Exception("gt path {} not working".format(gt_path))

            img_gt = cv2.cvtColor(img_gt, cv2.COLOR_BGR2RGB)
        else:
            try:
                img_gt = imfrombytes(img_bytes, flag='grayscale', float32=True)
            except:
                raise Exception("gt path {} not working".format(gt_path))

            img_gt = np.expand_dims(img_gt, axis=2)
        img_lq = img_gt.copy()


        # augmentation for training
        if self.opt['phase'] == 'train':
            gt_size = self.opt['gt_size']
            # padding
            img_gt, img_lq = padding(img_gt, img_lq, gt_size)

            # random crop
            img_gt, img_lq = paired_random_crop(img_gt, img_lq, gt_size, scale,
                                                gt_path)
            # flip, rotation
            if self.geometric_augs:
                img_gt, img_lq = random_augmentation(img_gt, img_lq)

            img_gt, img_lq = img2tensor([img_gt, img_lq],
                                        bgr2rgb=False,
                                        float32=True)


            if self.sigma_type == 'constant':
                sigma_value = self.sigma_range
            elif self.sigma_type == 'random':
                sigma_value = random.uniform(self.sigma_range[0], self.sigma_range[1])
            elif self.sigma_type == 'choice':
                sigma_value = random.choice(self.sigma_range)

            noise_level = sigma_value / 255.0
            # noise_level_map = torch.ones((1, img_lq.size(1), img_lq.size(2))).mul_(noise_level).float()
            # noise = np.random.randn(*img_lq.shape) * noise_level
            noise = paddle.randn(img_lq.shape,dtype='float32').numpy() * noise_level

            img_lq = img_lq + noise.astype('float32')

        else:
            np.random.seed(seed=0)
            img_lq += np.random.normal(0, self.sigma_test/255.0, img_lq.shape)
            # noise_level_map = torch.ones((1, img_lq.shape[0], img_lq.shape[1])).mul_(self.sigma_test/255.0).float()

            img_gt, img_lq = img2tensor([img_gt, img_lq],
                            bgr2rgb=False,
                            float32=True)

        return {
            'lq': img_lq,
            'gt': img_gt,
            'lq_path': gt_path,
            'gt_path': gt_path
        }

    def __len__(self):
        return len(self.paths)