Awesome

TBEFN: A Two-branch Exposure-fusion Network for Low-light Image Enhancement

Codes for TMM20 paper "TBEFN: A Two-branch Exposure-fusion Network for Low-light Image Enhancement".

Structure

requirements TensorFlow 1.x

tensorflow==1.13.1
opencv-python

requirements TensorFlow 2.x

tensorflow==2.6.0
tf-slim==1.1.0
opencv-python

get started

file structure

file	description
./input_dir	put your test image here
./results	output enhanced images
./ckpt	model weights (already provided, ~2MB)
./demo_img	used for demo

how to run the code
TensorFlow 1.x:

cd your_path
python predict_TBEFN.py

TensorFlow 2.x:

cd your_path
python predict_TBEFN_tf2.py

Colab

A Colab notebook which allows upload of your own photos and make predictions over them is available in this repository.

.pb file extension

See ./extension. First run TBEFN_ckpt2pb.py, and then TBEFN_RunFromPb.py.

other extensions

We thank PINTO0309's warm work that converted TBEFN into many other formats and for other platforms, including saved_model, tflite, onnx, coreml, tfjs, tftrt, openvino, myriad blob, edgetpu etc.

results

We provide 6 images in this demo, after running this code, you will get results as follows. (we have cropped the result so that you can have a better comparison.)

demo_img

further comparison

comparison with some other sota work (DEC.19)

demo_img

PSNR/SSIM/NIQE on paired dataset

demo_img

NIQE on six commonly used dataset

demo_img

Efficiency

demo_img

license

BSD 3-Clause

citation

@ARTICLE{lu2020tbefn,
  author={Lu, Kun and Zhang, Lihong},
  journal={IEEE Transactions on Multimedia}, 
  title={TBEFN: A Two-Branch Exposure-Fusion Network for Low-Light Image Enhancement}, 
  year={2021},
  volume={23},
  number={},
  pages={4093-4105},
  doi={10.1109/TMM.2020.3037526}}