tk-yoshimura/JSAI2021_YHSRNv7

Homogeneous responsive neural networks, YHSRNv7 (Oral presentation at JSAI 2021)

YHSRNv7

Yamatani-based Homogeneous Super Resolution Network v7 @ JSAI2021

Licence

CC BY-NC-ND

Author

tk-yoshimura

Report

J-STAGE
SlideShare(Japanese)
SlideShare(English)
preprint

Usage

Tensorflow 2.1.0
CUDA 11.*

Train Procedure

1. Generate "RandomPattern" images.
   Run InfinityPatterns, InfinityPatternsGenerator(directories:512 images:8192)

2. Make blur from "RandomPattern".
   Run makeblur/makebulr.py

3. Train and Sample SR model "YHSRNv7".
   Run YHSRNv7/train.py

4. Score SR.
   Run DIV2K_MATLAB_scoring eval_div2k_yhsrnv7.m

Validate Procedure

1. Sample SR model "YHSRNv7".
   Run YHSRNv7/sample.py

2. Score SR.
   Run DIV2K_MATLAB_scoring eval_div2k_yhsrnv7.m

Generate SR Image

Run YHSRNv7/sample.py

Sample