Uncertainty-Aware Unsupervised Domain Adaptation in Object Detection

Unsupervised domain adaptive object detection aims to adapt detectors from a labelled source an unlabelled target domain. Most existing works take two-stage strategy that first generates region proposals and then detects objects of interest, where adversarial learning is widely adopted mitigate the inter-domain discrepancy in both stages. However, may impair alignment well-aligned samples as it merely aligns global distributions across domains. To address this issue, we design uncertainty-aware adaptation network (UaDAN) introduces conditional align poorly-aligned separately different manners. Specifically, uncertainty metric assesses each sample adjusts strength for adaptively. In addition, exploit achieve curriculum performs easier image-level more difficult instance-level progressively. Extensive experiments over four challenging datasets show UaDAN achieves superior performance compared with state-of-the-art methods.