Skip to content

Latest commit

 

History

History
104 lines (73 loc) · 4.72 KB

README.md

File metadata and controls

104 lines (73 loc) · 4.72 KB
Raw

VDSR-PyTorch

Overview

This repository contains an op-for-op PyTorch reimplementation of Accurate Image Super-Resolution Using Very Deep Convolutional Networks.

Table of contents

About Accelerating the Super-Resolution Convolutional Neural Network

If you're new to VDSR, here's an abstract straight from the paper:

We present a highly accurate single-image superresolution (SR) method. Our method uses a very deep convolutional network inspired by VGG-net used for ImageNet classification. We find increasing our network depth shows a significant improvement in accuracy. Our finalmodel uses 20 weight layers. By cascading small filters many times in a deep network structure, contextual information over large image regions is exploited in an efficient way. With very deep networks, however, convergence speed becomes a critical issue during training. We propose a simple yet effective training procedure. We learn residuals onlyb and use extremely high learning rates (104 times higher than SRCNN) enabled by adjustable gradient clipping. Our proposed method performs better than existing methods in accuracy and visual improvements in our results are easily noticeable.

Download weights

Download datasets

Contains T91, Set5, Set14, BSDS100 and BSDS200, etc.

Test

Modify the contents of the file as follows.

  • line 30: upscale_factor change to the magnification you need to enlarge.
  • line 32: mode change Set to valid mode.
  • line 65: model_path change weight address after training.

Train

Modify the contents of the file as follows.

  • line 30: upscale_factor change to the magnification you need to enlarge.
  • line 32: mode change Set to train mode.

If you want to load weights that you've trained before, modify the contents of the file as follows.

  • line 47: start_epoch change number of training iterations in the previous round.
  • line 48: resume change weight address that needs to be loaded.

Result

Source of original paper results: https://arxiv.org/pdf/1511.04587.pdf

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

Dataset Scale PSNR
Set5 2 37.53(37.41)
Set5 3 33.66(33.44)
Set5 4 31.35(31.05)

Low Resolution / Super Resolution / High Resolution

Credit

Accurate Image Super-Resolution Using Very Deep Convolutional Networks

Jiwon Kim, Jung Kwon Lee, Kyoung Mu Lee

Abstract
We present a highly accurate single-image superresolution (SR) method. Our method uses a very deep convolutional network inspired by VGG-net used for ImageNet classification. We find increasing our network depth shows a significant improvement in accuracy. Our finalmodel uses 20 weight layers. By cascading small filters many times in a deep network structure, contextual information over large image regions is exploited in an efficient way. With very deep networks, however, convergence speed becomes a critical issue during training. We propose a simple yet effective training procedure. We learn residuals onlyb and use extremely high learning rates (104 times higher than SRCNN) enabled by adjustable gradient clipping. Our proposed method performs better than existing methods in accuracy and visual improvements in our results are easily noticeable.

[Paper] [Author's implements(MATLAB)]

@inproceedings{vedaldi15matconvnet,
  author    = {A. Vedaldi and K. Lenc},
  title     = {MatConvNet -- Convolutional Neural Networks for MATLAB},
  booktitle = {Proceeding of the {ACM} Int. Conf. on Multimedia},
  year      = {2015},
}