This repository is the official Tensorflow implementation for our unpublished paper：
EnhancedNet, an end-to-end network for dense disparity estimation and its application to aerial images
and our published paper
Context pyramidal network for stereo matching regularized by disparity gradients
The code is implemented with Python(3.8) and Tensorflow(1.15.5) for CUDA Version 12.0
- Donwload the pre-trained model and put it into the 'pre-trained' folder
pre-trained model donwload link: googledrive
- Run the inference.py with
python inference.py --left_path your_left_image_path --right_path your right_image_path --pretrain_model your_pretrain_model_path --net_type <initial or enhanced> --save_dir your_save_dir
- The resulting disparity maps are written to the save folder.
Please cite our paper if you use this code or any of the models:
@article{kang2019context,
title={Context pyramidal network for stereo matching regularized by disparity gradients},
author={Kang, Junhua and Chen, Lin and Deng, Fei and Heipke, Christian},
journal={ISPRS Journal of Photogrammetry and Remote Sensing},
volume={157},
pages={201--215},
year={2019},
publisher={Elsevier}
}