image super resolution

enviornment

• tensorflow
• ubuntu 18.04

data

• create a dir name data and put the training data in here.
• If you want to generate more data for data augmentation, run python3 augmentation.py --dataset="your dataset dir" --augment_level="you can choose 4/8".

train

• First, go to arg.py to set the parameter you want, like scale, lr, number of filter... and start training.

inference

• After training, you must use the same parameter to test the model, run python3 sr.py --file="your image path" and you can get upscale image for the low resolution image in the output dir.
• It will genertate 6 images for each lr image, the image named result is the upscale one.

result

low resolution image

high resolution image (x3)

reference

• https://github.com/jiny2001/dcscn-super-resolution