eshibusawa/jbf-stereo

GPU implementation of disparity refinement filter

JBF-Stereo

JBF-Stereo is an implementation for disparity refinement by using joint bilateral filtering (JBF). The refinement filter consists of joint bilateral filtering and joint nearest filtering, and both filters are CUDA-based GPU implementation.

Result of disparity refinement by using Joint Bilateral Filter

Result of PatchMatch Stereo

see more detail for PatchMatch.

Result of ELAS (Efficient Large-Scale Stereo Matching)

see more detail for ELAS.

Result of TGV (Total Generalized Variation)

see more detail for TGV.

Requirements

JBF-Stereo requires the following libraries:

• cupy
• opencv
• scipy
• nose (only for testing)

pip install -r requirement.txt

Usage

# compute disparity using block matching and apply JBF, PM, ELAS
python Python/stereo_main.py

# compute disparity using TGV
# this demo requires CUDAToolkit and executable nvcc
python Python/stereo_main_nvcc.py

jbf-stereo's Issues

Publications about this repository

HI.
I would like to cite your work, but I can't find any publications named "JBF-Stereo" anywhere.
I also searched for "Eijiro SHIBUSAWA stereo matching" but I got no results.
And I was able to find a paper titled "Double-direction fisheye plane sweep stereo for road surface 3D reconstruction", but I could not download the pdf.
I would appreciate it if you could provide me with a pdf or name of your publication.

About speed

The project is really outstanding, but when I used this matching method, I found that the speed was really slow. Did it call cuda correctly