This repository is the souce code of SHR-Net which is accepted to IROS2021 (with RA-L). For commercial purposes, please contact the authors: duhz@tju.edu.cn. If you use SRH-Net in an academic work, please cite:
inproceedings{dusrhnet,
author = {Hongzhi Du, Yanyan Li, Yanbiao Sun, Jigui Zhu and Federico Tombari},
title = {SRH-Net: Stacked Recurrent Hourglass Network for Stereo Matching},
year = {2021},
booktitle = {arXiv preprint arXiv:2105.11587},
}
We suggest to create an Anaconda environment and install the dependencies:
conda create -y -n SRHNET python=3.6
conda activate SRHNET
pip install -r requirements.txt
Please download the SceneFLow dataset: "FlyingThings3D", "Driving" and "Monkaa" (clean pass and disparity files).
-mv all training images (totallty 29 folders) into ${your dataset PATH}/frames_cleanpass/TRAIN/
-mv all corresponding disparity files (totallty 29 folders) into ${your dataset PATH}/disparity/TRAIN/
-make sure the following 29 folders are included in the "${your dataset PATH}/disparity/TRAIN/" and "${your dataset PATH}/frames_cleanpass/TRAIN/":
15mm_focallength 35mm_focallength A a_rain_of_stones_x2 B C
eating_camera2_x2 eating_naked_camera2_x2 eating_x2 family_x2 flower_storm_augmented0_x2 flower_storm_augmented1_x2
flower_storm_x2 funnyworld_augmented0_x2 funnyworld_augmented1_x2 funnyworld_camera2_augmented0_x2 funnyworld_camera2_augmented1_x2 funnyworld_camera2_x2
funnyworld_x2 lonetree_augmented0_x2 lonetree_augmented1_x2 lonetree_difftex2_x2 lonetree_difftex_x2 lonetree_winter_x2
lonetree_x2 top_view_x2 treeflight_augmented0_x2 treeflight_augmented1_x2 treeflight_x2
download and extract kitti and kitti2015 datasets.
Revise parameter settings and run "myevalution.sh" and "predict.sh" for evaluation and prediction on the SceneFLow dataset and KITTI datasets. Note that the “crop_width” and “crop_height” must be multiple of 16, "max_disp" must be multiple of 4 (default: 192).
The repo provides the pretrained model for testing. Please extract the .zip file into SRHNet Folder and use the following command to test your stereo images.
python test_img.py --crop_height= image height\
--crop_width= image width\
--max_disp=192\
--leftimg='path/to/left/image'\
--rightimg='path/to/left/image'\
--resume='path/to/pretrained/model'
As an example, we also provide stereo images that can be tested by using the following command,
python test_img.py --crop_height=384\
--crop_width=1248\
--max_disp=192\
--leftimg='./demo/left12_10.png'\
--rightimg='./demo/right12_10.png'\
--resume='./finetune2_kitti2015_epoch_8.pth'