A simple implementation for T3A (NeurIPS 2021 Spotlight)
-
This is a simple PyTorch implementation for the paper
Test-Time Classifier Adjustment Module for Model-Agnostic Domain Generalization
. -
The official implementation is here
-
The source model is trained on CIFAR-10, the training code & checkpoints are from here
-
Try
python utility.py
to compare the utility of the source model and a T3A-based model.