/super-resolution

image super resolution demo, from training to inference

Primary LanguagePython

super-resolution

model zoos

non-deeplearning

  • SCSR
  • SelfExSR
  • RFL

deeplearning

data

  • CUBIC(SRCNN) ./dataset_make/make_dataset.py
  • DIV2K / Flickr2K / BSD / ...
  • TextZoom

model

  • SRCNN / FSRCNN ./srresnet/
  • SRResNet / SRGAN ./srresnet/
  • VDSR / EDSR / MDSR ./edsr/
  • DBPN
  • RCAN ./rcan/
  • TTSR ./ttsr/ # NOTE(JiaKui Hu): rethinking
  • USRNet ./usrnet # NOTE(JiaKui Hu): rethinking
  • RFDN ./rfdn
  • ZSSR ./zssr
  • AcNet ./acnet
  • PAN ./pan
  • EPSANet

loss

  • L1/L2 ls
  • Perceptual loss ./srresnet/
  • Charbonnier loss ./loss.py
  • sinkhorn loss ./loss.py with L1 & Charbonnier
  • sinkhorn loss in edges

inference

  • TensorRT
  • TVM
  • vitis-ai

Usage

to make sure all the models could be run on your machine

pytest test.py

please run sudo apt install graphviz first if you open the --save_model_pdf

python train.py \
    --model_name ttsr \
    --train_file ./dataset_make/train_data.h5 \
    --valid_file ./dataset_make/valid_data.h5 \

or see more attrs in config.py