Skip to content

Latest commit

 

History

History
 
 

real-esrgan

README.md

Real-ESRGAN

The Pytorch implementation is real-esrgan.

Config

  • Input shape(INPUT_H, INPUT_W, INPUT_C) defined in real-esrgan.cpp
  • GPU id(DEVICE) can be selected by the macro in real-esrgan.cpp
  • BATCH_SIZE can be selected by the macro in real-esrgan.cpp
  • FP16/FP32 can be selected by PRECISION_MODE in real-esrgan.cpp
  • The example result can be visualized by VISUALIZATION.

How to Run, real-esrgan as example

  1. prepare test image
cd {tensorrtx}/real-esrgan/
mkdir sample   
cp ~/Download/OST_009.png {tensorrtx}/real-esrgan/sample
  1. generate .wts from pytorch with .pt, or download .wts from model zoo
git clone https://github.com/xinntao/Real-ESRGAN.git
cd Real-ESRGAN
pip install basicsr
pip install facexlib
pip install gfpgan
pip install -r requirements.txt
python setup.py develop

// download https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.0/RealESRGAN_x4plus.pth
cp ~/RealESRGAN_x4plus.pth {xinntao}/Real-ESRGAN/experiments/pretrained_models

cp {tensorrtx}/Real-ESRGAN/gen_wts.py {xinntao}/Real-ESRGAN
cd {xinntao}/Real-ESRGAN
python gen_wts.py
// a file 'real-esrgan.wts' will be generated.
  1. build tensorrtx/real-esrgan and run
cd {tensorrtx}/real-esrgan/
mkdir build
cd build
cp {xinntao}/Real-ESRGAN/real-esrgan.wts {tensorrtx}/real-esrgan/build
cmake ..
make
sudo ./real-esrgan -s [.wts] [.engine]   // serialize model to plan file
sudo ./real-esrgan -d [.engine] [image folder]  // deserialize and run inference, the images in [image folder] will be processed.
// For example
// sudo ./real-esrgan -s ./real-esrgan.wts ./real-esrgan_f32.engine
// sudo ./real-esrgan -d ./real-esrgan_f32.engine ../samples
  1. check the images generated, as follows. _OST_009.png