BayerUnifyAug

Python toolbox for Bayer Raw image unification and augmentation. This repo implements the data pre-processing part of our NTIRE 2019 Real Image Denoising Challenge submission. For more details, see our paper Learning Raw Image Denoising With Bayer Pattern Unification and Bayer Preserving Augmentation

Dependency

Python3, Numpy

Note: For Python 2, change the function signatures to Python 2 style.

How to Install

Copy bayer_unify_aug.py and you are good to go.

Demo

import bayer_unify_aug
import cv2

# load RGGB image
rggb = cv2.imread("./demo/lena_rggb.png", -1)

# output demosaic result
bgr = cv2.cvtColor(rggb, cv2.COLOR_BAYER_BG2BGR)
cv2.imwrite("./demo/lena_direct_out.jpg", bgr)

# convert RGGB to BGGR and then demosaic
bggr = bayer_unify_aug.bayer_unify(
    rggb, input_pattern="RGGB", target_pattern="BGGR", mode="pad")
bgr = cv2.cvtColor(bggr, cv2.COLOR_BAYER_RG2BGR)
cv2.imwrite("./demo/lena_bggr_out.jpg", bgr)

# flip RGGB image and then demosaic
flipped = bayer_unify_aug.bayer_aug(
    rggb, flip_h=False, flip_w=True, transpose=False, input_pattern="RGGB")
bgr = cv2.cvtColor(flipped, cv2.COLOR_BAYER_BG2BGR)
cv2.imwrite("./demo/lena_flipped_out.jpg", bgr)

Citing

If BayerUnify and BayerAug help your research, please cite it using the following BibTex entry:

@InProceedings{Liu_2019_CVPR_Workshops,
author = {Liu, Jiaming and Wu, Chi-Hao and Wang, Yuzhi and Xu, Qin and Zhou, Yuqian and Huang, Haibin and Wang, Chuan and Cai, Shaofan and Ding, Yifan and Fan, Haoqiang and Wang, Jue},
title = {Learning Raw Image Denoising With Bayer Pattern Unification and Bayer Preserving Augmentation},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2019}
}