This repository contains an op-for-op PyTorch reimplementation of Perceptual Extreme Super Resolution Network with Receptive Field Block.

• RFB_ESRGAN-PyTorch
  • Overview
  • Table of contents
  • Download weights
  • Download datasets
  • How Test and Train
    • Test
    • Train RFBESRNet model
    • Resume train RFBESRNet model
    • Train ESRGAN model
    • Resume train ESRGAN model
  • Result
  • Contributing
  • Credit
    • Perceptual Extreme Super Resolution Network with Receptive Field Block

• Google Driver
• Baidu Driver

Contains DIV2K, DIV8K, Flickr2K, OST, T91, Set5, Set14, BSDS100 and BSDS200, etc.

• Google Driver
• Baidu Driver

Both training and testing only need to modify the config.py file.

• line 31: upscale_factor change to 16.
• line 33: mode change to valid.
• line 111: model_path change to results/pretrained_models/RFBESRNet_x16-DFO2K-0bcd554c.pth.tar.

• line 31: upscale_factor change to 16.
• line 33: mode change to train_rfbesrnet.
• line 35: exp_name change to RFBESRNet_baseline.

• line 31: upscale_factor change to 16.
• line 33: mode change to train_rfbesrnet.
• line 35: exp_name change to RFBESRNet_baseline.
• line 49: resume change to samples/RFBESRNet_baseline/g_epoch_xxx.pth.tar.

• line 31: upscale_factor change to 16.
• line 33: mode change to train_rfbesrgan.
• line 35: exp_name change to RFBESRGAN_baseline.
• line 77: resume change to results/RFBESRNet_baseline/g_last.pth.tar.

• line 31: upscale_factor change to 16.
• line 33: mode change to train_rfbesrgan.
• line 35: exp_name change to RFBESRGAN_baseline.
• line 77: resume change to results/RFBESRNet_baseline/g_last.pth.tar.
• line 78: resume_d change to samples/RFBESRGAN_baseline/g_epoch_xxx.pth.tar.
• line 79: resume_g change to samples/RFBESRGAN_baseline/g_epoch_xxx.pth.tar.

Source of original paper results: https://arxiv.org/pdf/2005.12597v1.pdf

In the following table, the value in () indicates the result of the project, and - indicates no test.

Low resolution / Recovered High Resolution / Ground Truth

If you find a bug, create a GitHub issue, or even better, submit a pull request. Similarly, if you have questions, simply post them as GitHub issues.

I look forward to seeing what the community does with these models!

Taizhang Shang, Qiuju Dai, Shengchen Zhu, Tong Yang, Yandong Guo

Abstract
Perceptual Extreme Super-Resolution for single image is extremely difficult, because the texture details of different images vary greatly. To tackle this difficulty, we develop a super resolution network with receptive field block based on Enhanced SRGAN. We call our network RFB-ESRGAN. The key contributions are listed as follows. First, for the purpose of extracting multi-scale information and enhance the feature discriminability, we applied receptive field block (RFB) to super resolution. RFB has achieved competitive results in object detection and classification. Second, instead of using large convolution kernels in multi-scale receptive field block, several small kernels are used in RFB, which makes us be able to extract detailed features and reduce the computation complexity. Third, we alternately use different upsampling methods in the upsampling stage to reduce the high computation complexity and still remain satisfactory performance. Fourth, we use the ensemble of 10 models of different iteration to improve the robustness of model and reduce the noise introduced by each individual model. Our experimental results show the superior performance of RFB-ESRGAN. According to the preliminary results of NTIRE 2020 Perceptual Extreme Super-Resolution Challenge, our solution ranks first among all the participants.

[Paper]

@misc{2005.12597,
    Author = {Taizhang Shang and Qiuju Dai and Shengchen Zhu and Tong Yang and Yandong Guo},
    Title = {Perceptual Extreme Super Resolution Network with Receptive Field Block},
    Year = {2020},
    Eprint = {arXiv:2005.12597},
}