Unsupervised BatchNorm Adaptation (UBNA): A Domain Adaptation Method for Semantic Segmentation Without Using Source Domain Representations

Abstract

In this paper we present a solution to the task of "unsupervised domain adaptation (UDA) of a pre-trained semantic segmentation model without relying on any source domain representations". Previous UDA approaches for semantic segmentation either employed simultaneous training of the model in the source and target domains, or they relied on a generator network, replaying source domain data to the model during adaptation. In contrast, we present our novel Unsupervised BatchNorm Adaptation (UBNA) method, which adapts a pre-trained model to an unseen target domain without using-beyond the existing model parameters from pre-training-any source domain representations (neither data, nor generators) and which can also be applied in an online setting or using just a few unlabeled images from the target domain in a few-shot manner. Specifically, we partially adapt the normalization layer statistics to the target domain using an exponentially decaying momentum factor, thereby mixing the statistics from both domains. By evaluation on standard UDA benchmarks for semantic segmentation we show that this is superior to a model without adaptation and to baseline approaches using statistics from the target domain only. Compared to standard UDA approaches we report a trade-off between performance and usage of source domain representations. 1

Towards Unsupervised Online Domain Adaptation for Semantic Segmentation

“…Source-Free UDA. While online UDA does not prohibit the utilization of source data, we voluntarily waive this privilege likewise other source-free methods [22,28]. Consequently, our method is also applicable to scenarios when source data is private.…”

Transductive transfer learning for visual recognition

“…Our method transforms images into frequency space and divides images into DIFs from source domains to target domains to mitigate domain gaps [7,14,133,137,138,[178][179][180]. The third is self-training-based which employs "pseudo labels" to guide unsupervised learning over unlabeled target-domain data [13,141,143,[181][182][183][184].…”