Skip to content

Latest commit

 

History

History
14 lines (8 loc) · 481 Bytes

README.md

File metadata and controls

14 lines (8 loc) · 481 Bytes

ConvNeXt

This repository contains my custom implementation of ConvNeXt from scratch. It was originally developed for training on the VGGFace2 dataset.

My intention was to create a clean, easily understandable implementation of the network. If you have any suggestions, or constructive criticism to offer, please do not hesitate to create an issue.

It has the following fundamental changes from the paper -

  1. Uses BatchNorm instead of LayerNorm
  2. Does not use LayerScale