Skip to content

Latest commit

 

History

History
14 lines (12 loc) · 756 Bytes

README.md

File metadata and controls

14 lines (12 loc) · 756 Bytes

FreezeGroupNormalization

PoC / experiment / performance comparison of freeze group normalization with other state-of-the-art normalization method

  • The implementation of FreezeGroupNormalization was based on keras and it was all run on colab.
  • Used another model record system wandb for collecting and visualizing performance.
  • Used papermill to control & execute notebooks like notebook as a function.

Folder Structure

  • src: implementation of FreezeGroupNormalization source code.
  • configs: parameters file of model.
  • train.ipynb: training script.
  • main.ipynb: main entry for setup & execute train.ipynb.

Result