Deep Learning in Video Games: Super-Resolution

A Keras based implementation of

• Single Image Super-Resolution using a RestNet (SISRRN)

Requeriments

• python
• TensorFlow or TensorFlow-gpu v1.14
• Keras v2.2.4
• Pillow
• Skimage
• Numpy

Executing

Test

Apply Super Resolution for image file:

python Test.py --input_dir ./img/img.jpg --output_dir ./out/ --model_dir ./model/model.h5

'-i', '--input_dir': Path for input image

'-o', '--output_dir': Path for Output image

'-m', '--model_dir': Path for model

Train

If you want to train a new model with your own dataset, you can run:

python Train.py --input_dir ./data/ --output_dir ./out/ --model_dir ./model/ --epochs 5 --batch_size 64 --scale 2

'-i', '--input_dir': Path for input images

'-o', '--output_dir': Path for Output images

'-m', '--model_dir': Path for model

'-e', '--epochs', action='store': Number of epochs for train

'-b', '--batch_size', action='store': Batch size for train

'-s', '--scale': Upsampling scale (2 == x2, 4 == x4, ...)

Model

Result

Author

Imar Abreu Díaz - Computer Engineering Student

License

This project is licensed under the MIT License