FSRNet: End-to-End Learning Face Super-Resolution with Facial Priors

Tensorflow implement FSRNet based on SRN-Deblur

Testing

Download pretrained models and unzip, make sure the model path is ./checkpoints/color/checkpoints/deblur.model*

--input_path=<TEST_FOLDER> and save the outputs to --output_path=<OUTPUT_FOLDER>. For example:

python run_model.py --input_path=./testing_set --output_path=./testing_res --gpu=0 --model=color --phase=test --height=128 --width=128

Training

1. use data_loader.py to generate tfrecords in main function
2. Hyper parameters such as batch size, learning rate, epoch number can be tuned through command line:

python run_model.py --phase=train --batch=16 --lr=1e-4 --epoch=500

Some problems

1. Since the author do not open the code of cropping the face, so the dataset i use is different from theirs, our face is bigger than theirs.
2. I use face alignment to generate landmarks.
3. Download model from model password: 0z3l