Home

Awesome

DIP

Official PyTorch implementation of paper:

DIP: Deep Inverse Patchmatch for High-Resolution Optical Flow, CVPR 2022

Change Log

2023-12-12

upload stereo training codes, Mainly modified from unimatch and CRE-Pytorch

2023-10-19

Here we add stereo matching support for onnx

Installation

Our code is based on pytorch 1.6.0, CUDA 10.1 and python 3.8. The stereo matching version is based on pytorch 2.0.1, CUDA 11.7 and python 3.10.12

Demos

All pretrained models can be downloaded from google drive.

You can run a trained model on a sequence of images and visualize the results:

CUDA_VISIBLE_DEVICES=0 python demo.py \
--model DIP_sintel.pth

for stereo matching version:

CUDA_VISIBLE_DEVICES=0 python demo_stereo.py \
--model DIP_stereo.pth --path input_stereo_imgs

Datasets

The datasets used to train and evaluate DIP flow are as follows:

Furthermore, we supplemented the training data with stereo image pairs from the CRE dataset to enhance the diversity of scenarios for our stereo matching model.

Acknowledgements

This project is based on RAFT and SCV. We thank the original authors for their excellent work.