A pytorch implementation of SRCNN.
Original Paper: Image Super-Resolution Using Deep Convolutional Networks (ECCV 2014)
- Training data: T91 dataset
- Optimizer: Adam (learning rate = 0.00001)
- Number of iteration: 12,500,000 (2,500 epochs; 5,000 iterations per epoch)
- Chose best scoring model on validation.
- Validation dataset: Set5
| Scale | On Paper | Experiment | Difference |
|---|---|---|---|
| 2 | 36.66dB | 36.20dB | 0.46dB |
| 3 | 32.75dB | 32.44dB | 0.31dB |
| 4 | 30.49dB | 30.10dB | 0.39dB |
| x2 | x3 | x4 |
|---|---|---|
 |
 |
 |
- Python: 3.12
- CUDA: 12.2
- PyTorch Build: 2.5.1
$ pip install torch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 --index-url https://download.pytorch.org/whl/cu121
$ pip install -r requirments.txt# Quick inference (Images required in 'inference/input/'; Default scale: 3)
$ python inference.py
# Example usage
$ python inference.py -m 'pretrained_models/scale-4.BEST_PSNR.pth' -s 4 -i 'inference/input/' -o 'inference/results/'
# Usage
$ python inference.py [-m MODEL] [-s SCALE] [-i IMAGES] [-o OUTPUT]- Pretrained models are located at 'pretrained_models/'
# Example usage
$ python train.py --training_data '/data/T91/' --validataion_data '/data/Set5/'
# Usage
$ python train.py [--experiment_dir EXPERIMENT_DIR] [--scale_factor SCALE_FACTOR] [--learning_rate LEARNING_RATE] [--model_path MODEL_PATH] [--epochs EPOCHS] [--training_data TRAINING_DATA] [--validation_data VALIDATION_DATA]- Experiment results will be exported to 'experiments/' by default.
- Model will be fine-tuned after pretrained model if MODEL_PATH is given.