This project is an implementation of Image Super-Resolution Using Deep Convolutional Networks .
You can study the paper provided above and srcnn_report.pdf file to understand the underlying theoretical concepts.
- If you haven't setup cuda, you can download and install cuda toolkit and the cuda version that your NVIDIA-GPU supports.
- Download the Set 14 dataset .
- Download the DIV2K dataset .
git clone https://github.com/ChristosKonstantas/Super_Resolution_Convolutional_Neural_Network.gitpython -m venv venv
.\venv\Scripts\activate
pip install -r requirements.txtpython3 -m venv venv
source venv/bin/activate
pip install -r requirements.txtIn src folder
python train.py --train_dir "\path\to\DIV2K_train_HR" --eval_dir "\path\to\DIV2K_valid_HR" --output_dir "\path\to\output_dir"python infer.py --model_path "..\pretrained\model.pth" --image_path "your\path\to\pngFile"to see how good your training performed given a good image and compare it versus its bicubic interpolated counterpart.
or
python sisr.py --train_dir "\path\to\DIV2K_train_HR" --eval_dir "\path\to\DIV2K_valid_HR" --first_image "D:\Datasets\DIV2K_train_HR\DIV2K_train_HR\0001.png"for Single Image Super Resolution (SISR).