Image Super-Resolution Using Very Deep Residual Channel Attention Networks

An implementation of RCAN described in the paper using tensorflow. Image Super-Resolution Using Very Deep Residual Channel Attention Networks

Published in ECCV 2018, written by Y. Zhang, K. Li, L. Wang, B. Zhong, and Y. Fu

Requirement

• Python 3.6.5
• Tensorflow 1.13.1
• Pillow 6.0.0
• numpy 1.15.0
• scikit-image 0.15.0

Datasets

• DIV2K
• Benchmarks

Pre-trained model

• RCAN_X2
• RCAN_X3
• RCAN_X4

Train using your own dataset

python main.py --train_GT_path ./GT_path --train_LR_path ./LR_path --test_GT_path ./test_GT_path --test_LR_path ./test_LR_path --test_with_train True --scale 2(or 3, 4, ...) --log_freq 1000

• LR image and HR image pair should have same index when they are sorted by name respectively.
• You can refer to the script file (run.sh) in my repository

Test using benchmarks

1. Download pre-trained model.

• RCAN_X2
• RCAN_X3
• RCAN_X4

2. Unzip the pre-trained model file

tar -cvf model.tar

3. Test using benchmarks

python main.py --mode test --pre_trained_model ./model/RCAN_X2(or 3, 4) --test_LR_path ./benchmark_LR_path --test_GT_path ./benchmark_GT_path --scale 2(or 3, 4) --self_ensemble False

If you want to use self_ensemble, --self_ensemble option to True

• You can refer to the script file (run.sh) in my repository

Inference your own images

1. Download pre-trained model.

• RCAN_X2
• RCAN_X3
• RCAN_X4

2. Unzip the pre-trained model file

tar -cvf model.tar

3. Inference your own images

python main.py --mode test_only --pre_trained_model ./model/RCAN_X2(or 3, 4) --test_LR_path ./your_own_images --scale 2(or 3, 4) --chop_forward False

If your images are too large, OOM error can occur. In that case, --chop_forward option to True

Experimental Results

Qunatitative Results

Method	Scale	Set5	Set14	B100	Urban100
Bicubic	X2	33.66 / 0.9299	30.24 / 0.8688	29.56 / 0.8431	26.88 / 0.8403
RDN	X2	38.24 / 0.9614	34.01 / 0.9212	32.34 / 0.9017	32.89 / 0.9353
RCAN(paper)	X2	38.27 / 0.9614	34.12 / 0.9216	32.41 / 0.9027	33.34 / 0.9384
RCAN(my results)	X2	38.25 / 0.9615	34.07 / 0.9216	32.36 / 0.9020	33.12 / 0.9367

Method	Scale	Set5	Set14	B100	Urban100
Bicubic	X3	30.39 / 0.8682	27.55 / 0.7742	27.21 / 0.7385	24.46 / 0.7349
RDN	X3	34.71 / 0.9296	30.57 / 0.8468	29.26 / 0.8093	28.80 / 0.8653
RCAN(paper)	X3	34.74 / 0.9299	30.65 / 0.8482	29.32 / 0.8111	29.09 / 0.8702
RCAN(my results)	X3	34.75 / 0.9302	30.61 / 0.8470	29.31 / 0.8105	29.03 / 0.8693

Method	Scale	Set5	Set14	B100	Urban100
Bicubic	X4	28.42 / 0.8104	26.00 / 0.7027	25.96 / 0.6675	23.14 / 0.6577
RDN	X4	32.47 / 0.8990	28.81 / 0.7871	27.72 / 0.7419	26.61 / 0.8028
RCAN(paper)	X4	32.63 / 0.9002	28.87 / 0.7889	27.77 / 0.7436	26.82 / 0.8087
RCAN(my results)	X4	32.56 / 0.8996	28.89 / 0.7891	27.78 / 0.7434	26.81 / 0.8079

Qualitative results are will be updated soon!

Comments

If you have any questions or comments on my codes, please email to me. son1113@snu.ac.kr

Reference

[1] https://github.com/yulunzhang/RCAN