coded-dual-pixels: Coded Aperture Dual-Pixel Sensing (CADS)

Official Repository for Ghanekar et al. 2024 "Passive Snapshot Coded Aperture Dual-Pixel RGB-D Imaging"

Requirements

This code was tested on python=3.8. Required libraries: imageio=2.33.1, opencv=4.5.5, scikit-image=0.19.2, einops=0.8.0, lpips=0.1.4, torch=2.3.1+cu118, torchvision=0.18.1+cu118, torchmetrics=1.4.0

To set up the conda environment required for this project, follow these steps:

git clone https://github.com/shadowfax11/coded-dual-pixels.git
cd coded-dual-pixels
conda env create -f environment.yml
conda activate cads

Usage

For training, run

sh train.sh <GPU_ID> <DATA_DIR_FOR_FLYINGTHINGS3D_DATASET>

For testing, run

sh test.sh <GPU_ID> <DATA_DIR_FOR_FLYINGTHINGS3D_DATASET>

Updates

2024-06-12: Code repo is updated and functioning

References/Credit

Some of the code functions, formatting is taken from other GitHub repos:

1. https://github.com/Abdullah-Abuolaim/recurrent-defocus-deblurring-synth-dual-pixel
2. https://github.com/ChenyangLEI/sfp-wild