# CGI-Stereo

**Repository Path**: ykuo/cgi-stereo

## Basic Information

- **Project Name**: CGI-Stereo
- **Description**: 复现先进算法：CGI-Stereo
复现代码来源于开源代码：gangweiX/CGI-Stereo
复现论文：CGI-Stereo: Accurate and Real-Time Stereo Matching via Context and Geometry Interaction
- **Primary Language**: Python
- **License**: MIT
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2024-01-06
- **Last Updated**: 2025-08-16

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# CGI-Stereo

#### 介绍

- 复现先进算法：CGI-Stereo
- 复现代码来源于开源代码：[gangweiX/CGI-Stereo](https://github.com/gangweiX/CGI-Stereo)
- 复现论文：CGI-Stereo: Accurate and Real-Time Stereo Matching via Context and Geometry Interaction

#### 软件架构

- python3.11
- pytorch2.1.0+cu121

#### 安装教程

1.  xxxx
2.  xxxx
3.  xxxx

#### 使用说明

1.  tensorboard --logdir=./logs --port=6006
2.  watch -n 1 nvidia-smi
3.  netstat -ano | findstr :6006
4.  nohup python ./main.py > ./logs/nohup.log 2>&1 &
5.  tail -f ./logs/nohup.log
6.  ps aux|grep python

#### 训练Training
    # Train CGI-Stereo on Scene Flow. First training,
    python train_sceneflow.py --saveckpt ./checkpoints/sceneflow/first/
    # Second training,
    python train_sceneflow.py --saveckpt ./checkpoints/sceneflow/second/ --loadckpt ./checkpoints/sceneflow/first/first.ckpt

#### 微调Finetune
    # Finetune CGI-Stereo on KITTI (using pretrained model on Scene Flow),
    python -u train_CGI_Stereo.py --batch_size 12 --test_batch_size 10 --num_workers 12
    # Finetune CGI-Stereo on USVInland (using pretrained model on Scene Flow),
    python -u train_CGI_Stereo.py --dataset usvinland --maxdisp 64 --trainlist ./filenames/usvinland_train.txt --testlist ./filenames/usvinland_val.txt --batch_size 36 --test_batch_size 36 --num_workers 18
    # Finetune CGI-Stereo on USVInland_seg (using pretrained model on Scene Flow),
    python -u train_CGI_Stereo.py --dataset usvinland_seg --maxdisp 64 --trainlist ./filenames/usvinland_seg_train.txt --testlist ./filenames/usvinland_seg_val.txt --batch_size 36 --test_batch_size 36 --num_workers 18

    # Finetune CGF-ACV on KITTI (using pretrained model on Scene Flow),
    python -u train_CGF_ACV.py --batch_size 2 --test_batch_size 2 --num_workers 10
    # Finetune CGI-Stereo on USVInland (using pretrained model on Scene Flow),
    python -u train_CGF_ACV.py --dataset usvinland --maxdisp 64 --trainlist ./filenames/usvinland_train.txt --testlist ./filenames/usvinland_val.txt --batch_size 12 --test_batch_size 12 --num_workers 12
    # Finetune CGI-Stereo on USVInland_seg (using pretrained model on Scene Flow),
    python -u train_CGF_ACV.py --dataset usvinland_seg --maxdisp 64 --trainlist ./filenames/usvinland_seg_train.txt --testlist ./filenames/usvinland_seg_val.txt --batch_size 12 --test_batch_size 12 --num_workers 12

#### 验证Evaluation
    # Generate disparity images of KITTI test set,
    python save_disp.py

#### 参与贡献

1.  Fork 本仓库
2.  新建 Feat_xxx 分支
3.  提交代码
4.  新建 Pull Request

#### 特技

1.  使用 Readme\_XXX.md 来支持不同的语言，例如 Readme\_en.md, Readme\_zh.md
2.  Gitee 官方博客 [blog.gitee.com](https://blog.gitee.com)
3.  你可以 [https://gitee.com/explore](https://gitee.com/explore) 这个地址来了解 Gitee 上的优秀开源项目
4.  [GVP](https://gitee.com/gvp) 全称是 Gitee 最有价值开源项目，是综合评定出的优秀开源项目
5.  Gitee 官方提供的使用手册 [https://gitee.com/help](https://gitee.com/help)
6.  Gitee 封面人物是一档用来展示 Gitee 会员风采的栏目 [https://gitee.com/gitee-stars/](https://gitee.com/gitee-stars/)

#### 引用
> @inproceedings{xu2023cgi,
  title={CGI-Stereo: Accurate and Real-Time Stereo Matching via Context and Geometry Interaction},
  author={Xu, Gangwei and Zhou, Huan and Yang, Xin},
  journal={arXiv preprint arXiv:2301.02789},
  year={2023}
}