VRDL_HW4: Super Resolution

This is homework 4 in NYCU Selected Topics in Visual Recognition using Deep Learning.

Installation

build the environment via:

$ conda env create -f environment.yml

And install following packages:

$ pip install imageio
$ pip install opencv-python
$ pip install pandas
$ pip install flags
$ pip install scipy==1.2.2

Prepare Dataset

Unzip the given dataset and put it in vrdl_data folder The dataset folder should be like:

./vrdl_data
  |---training_hr_images/training_hr_images/*.png
  |---testing_lr_images/testing_lr_images/*.png

(For Training) For preparing training data, run the below code to genearate more training samples. (Please make sure that the data directory is same as above form)

$ python Prepare_TrainData_HR_LR_VRDL.py
$ python Prepare_val_data.py

Training Code

Start training with following command:

$ python train.py -opt ./options/train/train_SRFBN_VRDL.json

Evaluation code

download the pretrained model bellow and put it at ./submission.pth
Run the testing code:

$ python test.py -opt ./options/test/test_SRFBN_VRDL.json

Download Pretrained Models

Here is the model weight of my final submission. Please download the weights and run the above evaluation code.

Final_submission_weights

Reference

My howework references the codes in the following repos. Thanks for thier works and sharing.

SRFBN