Pyramid Stereo Matching Network

This repository contains the code (in PyTorch) for "Pyramid Stereo Matching Network" paper (CVPR 2018) by Jia-Ren Chang and Yong-Sheng Chen.

original repo

https://github.com/JiaRenChang/PSMNet

difference

this repo estimate Left and Right Disparity(original is left only). so you can train without Ground Truth. It is more practical to use

Dependencies

tested environment

Python 3.69
PyTorch(1.5.0)
torchvision 0.6.0
KITTI Stereo
[CUDA 10.2] [https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu&target_version=1804&target_type=debnetwork]
[Ubuntu 18.04]

installation

sudo apt install python3-pip

pip3 install torch torchvision

pip3 install tensorboard

Train

sh train.sh

Inference

sh inference.sh

visualize training log by tensorboard

tensorboard --logdir {your_workingdir}/logs

access from your browser http://localhost:6006/