Stereo_3DReconstruction

3D Reconstruction from Stereo Images. Our code Computes the disparity using PSMNet Architecture Model and then triangulate the 3D Points. Finally, 3D Points are visualized using Open3D.

PSMNet=> " https://github.com/JiaRenChang/PSMNet "

CapStone Project

This project is for 2020-1 semester Capstone Project.

1. Contributor

• Jae Won Yang, Jae Won U

2. Version

• python3.6
• Open3d 0.10
• torch1.5 and torchvision 0.6
• Cuda >=10.1
• Ubuntu 18.04
• Multiple GPU: Nvidia Titan X (Use Dataparallel GPU Number 2 and 3)

3. Build

• git clone "https://github.com/YangJae96/Stereo_3DReconstruction.git"
• Run "pip3 install -r requirments.txt"

4. Run Demo

• Command: "python3 reconstruction.py --folder_name chair"

5. Add custom dataset

• Stereo Datasets => "https://vision.middlebury.edu/stereo/data/scenes2014/"
• Download the dataset zip file
• Create dir inside the dataset dir -> (The dir name will be an argument when you type in the command)
• Put the calib.txt, im0.png, im1.png inside the (Your dataset) dir.
• Then Run "python3 reconstruction.py --folder_name (Your data dir name inside dataset dir)"