image super resolution with upscaling factor=3
training dataset:291 HR images
testing dataset:14 LR images
- colab GPU(Tesla P100)
- pytorch
- numpy
- Pillow
- torchvision
- scikit-learn
- scikit-image
- opencv-python
- tensorboard
there are
- 10 files
1. datasets.py
define the custom dataset
2. eval.py
evaluate the validation set
3. main.ipynb
run training procedure
4. model.py
SRResnet architecture in this file
5. swinIR.py
implementation of swinLR, is fromthe official totorial of [4]
6. test.py
use model to do the image super resolution
7. train.py
implement training procedure,I also modified its code to fit our experiments
8. train_val_split.py
split dataset to training dataset and validation dataset
9. utils.py
something we need but not included in any py files above
10. inference.ipynb
reproduce the submission - 2 folders
1. data
the place where your testing data and training data is
2. models
This folder used to put the trained model weights
- you only need to download
inference.ipynband just run it on colab.
note
The model para you need to reproduce the submission are embedded in the inference.ipynb, but if you want to check my weight para is normal, please follow this link
[1] https://github.com/sgrvinod/a-PyTorch-Tutorial-to-Super-Resolution
[2] Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network.By C.Ledig et al.
[3] SwinIR: Image Restoration Using Swin Transformer.By J.Liang et al.
[4] https://github.com/jingyunliang/swinir
[5] A comprehensive review of deep learning based single image super-resolution. By Syed Muhammad Arsalan Bashir et al.
[6] Image Super-Resolution Using Deep Convolutional Networks. By Chao Dong et al.
[7] Deep learning based single image super-resolution: a survey. International Journal of Automation and Computing, by Ha et al.
[8] Enhanced Deep Residual Networks for Single Image Super-Resolution.By Bee Lim, et al.