Skip to content

Latest commit

 

History

History
18 lines (13 loc) · 948 Bytes

README.md

File metadata and controls

18 lines (13 loc) · 948 Bytes
Raw

AMC-Loss: Angular Margin Contrastive Loss for Improved Explainability in Image Classification

Overview

This is re-implementation of the AMC loss described in: paper

Requirements

  • tensorflow>=1.12.0
  • python>=3.6.0

Image Classification

We use CIFAR10 classification as an example with a simple architecture. In order to reproduce the results described on the paper, please modify the hyperparameters in Train_AMC.py and then run python Train_AMC.py The users can also change the data to other dataset at their interest. This code contains visualization by Grad-CAM from here [https://github.com/Ankush96/grad-cam.tensorflow] and tSNE plot.

Visualization

Baseline model AMC model