This repository is part of CS7180 Advanced Perception course at Northeastern University. This is Assignment-1, Image Sharpening aimed at exploring Super Resolution techniques. We implement SRCNN in pytorch from scratch and try to swap the MSE loss with perceptual loss.
Follow these steps to train on Div2k dataset.
git clone https://github.com/prajapatisarvesh/SRCNN-Pytorch.git # Clone this repo
cd SRCNN-Pytorch
cd utils
### Download the dataset and prepare the CSV
python3 div2k_downloader.py
cd ..
python3 train.py
This should start training, tensorboard is enabled, and checkpoints are being saved after every epoch. Argument Parsing is one thing we want to add some point in time.
Some baseline checkpoints are provided with this model, these are our baselines, and you can test them with color images of any dimension.
cd SRCNN-Pytorch
python3 test.py # This should save output in output dir.
Thanks to this amazing repository that taught us some good practices to work with Pytorch -- Pytorch-Template. This repo will have lot's of improvement, we will keep updating this.