Tent: Fully Test-time Adaptation by Entropy Minimization

TL;DR

Tent is a test-time adaptation method that minimizes prediction entropy to improve model generalization on new data, achieving state-of-the-art results across various tasks without retraining.

Contribution

It introduces a fully test-time adaptation approach using entropy minimization, estimating normalization and affine parameters online during testing.

Findings

Reduces error on corrupted ImageNet and CIFAR datasets.

Achieves state-of-the-art performance on ImageNet-C.

Effective in source-free domain adaptation scenarios.

Abstract

A model must adapt itself to generalize to new and different data during testing. In this setting of fully test-time adaptation the model has only the test data and its own parameters. We propose to adapt by test entropy minimization (tent): we optimize the model for confidence as measured by the entropy of its predictions. Our method estimates normalization statistics and optimizes channel-wise affine transformations to update online on each batch. Tent reduces generalization error for image classification on corrupted ImageNet and CIFAR-10/100 and reaches a new state-of-the-art error on ImageNet-C. Tent handles source-free domain adaptation on digit recognition from SVHN to MNIST/MNIST-M/USPS, on semantic segmentation from GTA to Cityscapes, and on the VisDA-C benchmark. These results are achieved in one epoch of test-time optimization without altering training.