Codes for TMM20 paper "TBEFN: A Two-branch Exposure-fusion Network for Low-light Image Enhancement".
tensorflow==1.13.1
opencv-python
tensorflow==2.6.0
tf-slim==1.1.0
opencv-python
- 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
A Colab notebook which allows upload of your own photos and make predictions over them is available in this repository.
See ./extension. First run TBEFN_ckpt2pb.py, and then TBEFN_RunFromPb.py.
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.
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.)
- comparison with some other sota work (DEC.19)
- PSNR/SSIM/NIQE on paired dataset
- NIQE on six commonly used dataset
- Efficiency
BSD 3-Clause
@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}}