Yinglong Li, Jiacheng Li, Zhiwei Xiong

CVPR Paper

Updating! Any questions, please contact me at any time.

Some pretrained LUTs and their compressed version can be download here.

Let's take the SPF-LUT for x4 sr as an example.

cd sr
python 1_train_model.py --model SPF_LUT_net --scale 4 --modes sdy --expDir ../models/spf_lut_x4 --trainDir ../data/DIV2K --valDir ../data/SRBenchmark

The trained LUT network will be available under the ../models/spf_lut_x4 directory.

python .\2_compress_lut_from_net.py --model SPF_LUT_net --scale 4 --modes sdy --expDir ../models/spf_lut_x4 --lutName spf_lut_x4 --cd xyzt --dw 2 --si 5

The compressed LUTs will be available under the ../models/spf_lut_x4 directory. --cd: The number of compressed dimensions; --dw: Diagonal width; --si: Sampling interval of non-diagonal subsampling.

python 3_finetune_lut.py --model SPF_LUT_DFC --scale 4 --modes sdy --expDir ../models/spf_lut_x4  --trainDir ../data/DIV2K --valDir ../data/SRBenchmark --load_lutName spf_lut_x4 --cd xyzt --dw 2 --si 5

The finetuned compressed LUTs will be available under the ../models/spf_lut_x4 directory.

python .\4_test_SPF-LUT_DFC.py --scale 4 --modes sdy --expDir ../models/spf_lut_x4 --testDir ../data/SRBenchmark --lutName weight --cd xyzt --dw 2 --si 5

If you have any questions, feel free to contact me any time by e-mail [email protected]

If you found our implementation useful, please consider citing our paper:

@InProceedings{Li_2024_CVPR, 
	author = {Li, Yinglong and Li, Jiacheng and Xiong, Zhiwei}, 
	title = {Look-Up Table Compression for Efficient Image Restoration}, 
	booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)}, 
	month = {June}, year = {2024}, pages = {26016-26025} 
}

This work is based on the following works, thank the authors a lot.

SR-LUT

MuLUT