pjwang317 / N_SR

Some implements of state-of-the-art Super-Resolution architectures with Pytorch.

N_SR

There are some implements of image super-resolution methods with Pytorch.

• AdderSR in addersr.py (AdderSR: Towards Energy Efficient Image Super-Resolution) (2021)
• SMSR in smsr.py (Learning Sparse Masks for Efficient Image Super-Resolution) (2021)
• HAN in han.py (Single Image Super-Resolution via a Holistic Attention Network) (2020)
• SAN in san.py (Second-order Attention Network for Single Image Super-Resolution) (2020)
• CSNLA in csnla.py (Image Super-Resolution with Cross-Scale Non-Local Attention and Exhaustive Self-Exemplars Mining) (2020)
• DRN in drn.py (Closed-loop Matters: Dual Regression Networks for Single Image Super-Resolution) (2020)
• RCAN in rcan.py (Image Super-Resolution Using Very Deep Residual Channel Attention Networks) (2019)
• DBPN in dbpn.py (Deep Back-Projection Networks For Super-Resolution) (2018)
• RDN in rdn.py (Residual Dense Network for Image SR) (2018)
• EDSR in edsr.py (Enhanced Deep Residual Networks for Single Image Super-Resolution) (2017)
• SRResNet in srresnet.py (Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network) (2016)
• SRCNN in srcnn.py (Image Super-Resolution Using Deep Convolutional Networks) (2014)

Dependencies

• Python 3.5.3 +
• Pytorch 1.1.0
• tensorboardX
• pytorch-colors

Getting started

Image-based algoritnms.

• Train: python main.py.
  
• Test: python test.py. More details in option.py.
  

Patch-based algoritnms.

• Image to patch:python image_to_patch.py.
  

• Train: python main.py.
  

• Test: python test.py. More details in option.py.
  

• tensorborad --logdir ./log
  

Plot the LAM map

• python ./LAM/LAM_main.py. (Interpreting Super-Resolution Networks with Local Attribution Maps)
  

Experiments on FEI face dateset (without augmentation and pre-train)

Image-based algoritnms.

Algorithms	PSNR
Bicubic	36.38
EDSR_original	39.81
EDSR+b16k64	39.85
EDSR+b32k256	40.05
SRResNet without BN	40.04
RDN	39.90
DBPN	40.14
RCAN	40.03
SAN	39.97
HAN	40.04
SMSR	39.86

Patch-based algoritnms.

Algorithms	Bicubic	SRCNN_original	SRCNN	VDSR_original
PSNR	36.38	38.58	38.61	39.54

License

This project is released under the Apache 2.0 license.