Skip to content

Latest commit

 

History

History
18 lines (11 loc) · 787 Bytes

README.md

File metadata and controls

18 lines (11 loc) · 787 Bytes

Acknowledgements

This is a wrapper for the EffNet-B0 by LukeMelas/EfficientNet-PyTorch Thank him for training the models! I just wanted to see LRP on a more modern net.

Quickstart

Copy files into the path of EfficientNet-PyTorch-master. Set the paths to imagenet validation data. Run efficientnettestcode2.py.

What it supports

It supports for conv layers: beta0, beta and adaptive beta. See the code for lrp_layer2method['nn.Conv2d']. You can code others for other layers. Uses conv-batchnorm canonization. This code is from July 2021. It works with Pytorch 1.10. Heatmaps tend to be more visually neat than with resnets.