https://github.com/bloc97/torch-vdsr

Torch implementation of the VDSR-CNN Upscaling algorithm
https://github.com/bloc97/torch-vdsr

cnn convolutional-neural-networks neural-network srcnn super-resolution torch torch-vdsr torch7 upscaling vdsr vdsr-algorithm

Last synced: 26 days ago
JSON representation

Torch implementation of the VDSR-CNN Upscaling algorithm

• Host: GitHub
• URL: https://github.com/bloc97/torch-vdsr
• Owner: bloc97
• License: mit
• Created: 2017-01-29T14:36:37.000Z (almost 8 years ago)
• Default Branch: master
• Last Pushed: 2017-02-16T19:45:44.000Z (almost 8 years ago)
• Last Synced: 2024-10-14T17:21:39.769Z (2 months ago)
• Topics: cnn, convolutional-neural-networks, neural-network, srcnn, super-resolution, torch, torch-vdsr, torch7, upscaling, vdsr, vdsr-algorithm
• Language: Lua
• Size: 2.39 MB
• Stars: 5
• Watchers: 4
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: License.txt

Awesome Lists containing this project

README

        
# Torch-VDSR

Torch implementation of the VDSR-CNN Upscaling algorithm

This is a simple torch implementation of the VDSR algorithm from this paper, though some hyperparameters were modified for faster convergence.  

[Jiwon Kim, Jung Kwon Lee and Kyoung Mu Lee, **Accurate Image Super-Resolution Using Very Deep Convolutional Networks**][1]

Included model uses only 8 layers (instead of the paper's 20), was trained for 2 hours on a R9 290x using 7 high-resolution images.  

This is a preliminary test so the results are not the best. When I have time I will train the full 20 layer model using a bigger dataset (around 1000 images).  

![miku comparison](https://github.com/bloc97/Torch-VDSR/raw/master/demo/miku_comp.png)

![bird comparison](https://github.com/bloc97/Torch-VDSR/raw/master/demo/bird_comp.png ) 

[1]: https://arxiv.org/abs/1511.04587