freedomtan/some_super_resolution_tflite_models

Some super resolutions models converted to TFLite

some_super_resolution_tflite_models

Some super resolutions models converted to TFLite

files

models/saved_model

TensorFlow models in saved_model format.

• models/saved_model/{srgan,edsr}: from Martin Krasser's super resolution repo

• models/saved_model/wdsr: from offical WDSR

• models/saved_model/esrgan: from TFHub ESRGAN

models/checkpoint/

Some checkpoints for ABPN

• models/checkpoint/abpn_x2: abpn x2 trained with DIV2K
• models/checkpoint/abpn_x2_ip: abpn x2 trained with ImagePairs

models/tflite

fp32 tflite models

models/tflite_quant

post-training full integer quantized tflite models

scripts used to train ABPN with ImagePairs

train_abpn_ip.py

scripts used to generate tflite models

srgan.py, edsr.py, wdsr.py, esrgan.py

Jupyter notebooks used to check tflite models

test srgan tflite.ipynb, test edsr tflite.ipynb, test wdsr tflite.ipynb, test esrgan tflite.ipynb

other files

data.py from Martin Krasser's super resolution repo, to use DIV2K dataset

A self-contained Jupyter notebook that can be used train ABPN

• PSNR, SSIM evaluated with DIV2K: 33.37, 0.93
• PSNR, SSIM evaluated with ImagePairs: 22.06, 0.73

train_abpn_ip.py: a self-contained script to train ABPN with ImagePairs dataset

• (PSNR, SSIM) evaluated with ImagePairs: (23.95284, 0.76988685)

demo/*.png: low resolution cropped png from Martin Krasser's super resolution repo