We implement a tensorflow model for "Accurate Image Super-Resolution Using Very Deep Convolutional Networks", CVPR 16'.
- The author's project page
- We use 291 dataset as training dataset.
- Ubuntu 16.04
- Python 3.5
- Numpy
- Opencv2
- matplotlib
- main.py : Execute train.py and pass the default value.
- vdsr.py : VDSR model definition.
- train.py : Train the VDSR model and represent the test set performance.
- demo.py : Test the VDSR model and show result images and psnr.
- util.py : Utility functions for this project.
- log.txt : The log of training process.
- model : The save files of the trained VDSR.
# !!!IMPORTANT!!!
# You should execute aug_train.m and aug_test.m in 'data' directory.
# Recommend 'Octave' platform to execute matlab code '.m'
aug_train.m
aug_test.m
python preprocess.pypython main.py
# if you want to change training epoch ex) 80 epoch (default) -> 120 epoch
python main.py --training_epoch 120python demo.py
# default args: image_index = 1, scale = 2, coordinate = [50,50], interval = 30
# you can change args: image_index = 13, scale = 4, coorindate [100,100], interval = 50
python demo.py --image_index 13 --scale 4 --coordinate [100,100] --interval 50| Scale | Bicubic | tf_VDSR |
|---|---|---|
| 2x - PSNR | 31.41 | 33.15 |
-
Original (Urban100 / index 1)
-
Bicubic (Urban100 / index 1)
-
VDSR (Urban100 / index 1)