This dataset is a synthetic dataset for intrinsic decompostion, which provides photorealistic rendered images with ground truth albedo images and shading images.
There are about 20K samples in this dataset and we split them into train/test splits, which are respectively listed in resized_data/train_list.txt and resized_data/val_list.txt.
In folder resized-data, we provide resized images used during the network training for lowering memory costs. These images are resized to a half of their original sizes. There are several sub-folders:
input-resize: this folder contains all of the input images.diffuse-resize: this folder contains all of the corresponding albedo images.shading-resize: this folder contains all of the corresponding shading images.mask-resize: this folder contains all of the corresponding mask images, which are calculated and used for masking out some areas during loss calculation, such as areas with self-illumination, refraction, etc.
In folder raw-data, we provide the original high resolution images, where the data is organized in a similar way. Due to the large data volume of the raw data, we will provide it later.
A number of examples from our dataset are shown below.
Please download the data from Baidu Netdisk or Dropbox with the links given in download_address.txt.
If you use this dataset for your research, please consider citing our paper:
@article{wang2022intrinsic,
author = {Wang, Yujie and Fan, Qingnan and Li, Kun and Chen, Dongdong and Yang, Jingyu and Lu, Jianzhi and Lischinski, Dani and Chen, Baoquan},
title = {High quality rendered dataset and non-local graph convolutional network for intrinsic image decomposition},
journal = {Journal of Image and Graphics},
volume = {27},
number = {2},
pages = {404--420},
year = {2022},
doi = {10.11834/jig.210705},
}.
中文论文引用方式：
王玉洁, 樊庆楠, 李坤, 陈冬冬, 杨敬钰, 卢健智, Dani Lischinski, 陈宝权. 2022.
面向本征图像分解的高质量渲染数据集与非局部卷积网络. 中国图象图形学报, 27(2): 404-420.
[DOI: 10.11834/jig.210705]
If you have any questions, please feel free to contact us via email: yujiew.cn@gmail.com.