InStereo2K Dataset contains 2050 pairs of images with high accuracy disparity maps (2000 for training, 50 for testing), which can be downloaded at the following website: https://github.com/YuhuaXu/StereoDataset
The directory structure should be:
InStereo2K
├── train
| ├── part1
| | ├── 000000
| | ├── 000001
...
| ├── part2
...
├── test
| ├── 000040
| ├── 000080
...
Optionally you can write your own txt file and use all the parts of the dataset.
@article{bao2020instereo2k,
title={Instereo2k: a large real dataset for stereo matching in indoor scenes},
author={Bao, Wei and Wang, Wei and Xu, Yuhua and Guo, Yulan and Hong, Siyu and Zhang, Xiaohu},
journal={Science China Information Sciences},
volume={63},
pages={1--11},
year={2020},
publisher={Springer}
}