Skip to content

PyTorch implementation of Deep Convolution Networks based on EDSR for Compression(Jpeg) Artifacts Reduction

Notifications You must be signed in to change notification settings

developer0hye/EDAR

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

42 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

EDAR

PyTorch implementation of Deep Convolution Network based on EDSR for Compression Artifacts Reduction

Requirements

  • PyTorch
  • tqdm
  • Pillow

Network Architecture

fig1_EDAR_EXAMPLE

fig2_EDAR

Visual Results

fig4_bettertomorrow2_better

fig4_bettertomorrow_better

fig4_goorinimage

fig4_bridge

fig4_iu

fig4_ronaldo

fig4_mpeg

fig4_navi

Training

Dataset: DIV 2K train set + ...(custom dataset...)

Batch size: 16

Patch size: 48x48

Optimizer: Adam

Loss: L1 Loss

Input: Compressed Image by JPEG (jpeg_quality: rand(0 to 10)) / RGB

Output: Original Image / RGB

Epoch: 450

Pre-trained weight

How to train

python train.py --images_dir [Your training image path] --outputs_dir ./ --jpeg_quality [10 to 100] --batch_size [num] --num_epochs [num]

Pre-trained model was trained using the below arguments.

python train.py --images_dir ../DIV2K_train_HR --outputs_dir ./ --jpeg_quality 10 --batch_size 16 --num_epochs 200

How to test

python test.py --weights_path [your trained weight].pth --image_path [your_image] --outputs_dir ./

About

PyTorch implementation of Deep Convolution Networks based on EDSR for Compression(Jpeg) Artifacts Reduction

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages