SRnet-Demo

May 3, 2021 ยท View on GitHub

File Test.py is the main app.

You will need to download pretrained weights for:

  1. The craft model: LINK
  2. The Tensorflow SRNET model: LINK
  3. VGG 19 tensorflow weights: LINK

Place these weights in their appropriate folders (look at the .py files for their paths)

Setup a tensorflow virtual environment (tf version 1.15)

Install all dependencies.

Run test.py in the virtual env.

Note:

  • This demo uses the tensorflow version of SRNET. Change the code to the pytorch version (on my github).
  • Streamlit is extremely slow because it has to load weights for the CRAFT model as well as for the SRNET model. Find a way to change this. Feel free to build from scratch.
  • You could perhaps write a script for the bounding box detection and remove the craft model.
  • I won't be able to help / assist because I'm extremely busy.

Please maintain this codebase however you see fit. Once you have a fast demo running, you can opensource it through this repo. You'll get all credit of course.