MeGA-CDA: Memory Guided Attention for Category-Aware Unsupervised Domain Adaptive Object Detection

TL;DR

This paper introduces MeGA-CDA, a novel method for unsupervised domain adaptive object detection that incorporates category information into feature alignment using memory-guided attention, leading to improved performance.

Contribution

It proposes a memory-guided attention mechanism to enable category-aware domain adaptation without requiring target category labels.

Findings

Outperforms existing domain adaptation methods on benchmark datasets.

Effectively generates category-specific attention maps for feature routing.

Enhances discriminative feature learning for object detection across domains.

Abstract

Existing approaches for unsupervised domain adaptive object detection perform feature alignment via adversarial training. While these methods achieve reasonable improvements in performance, they typically perform category-agnostic domain alignment, thereby resulting in negative transfer of features. To overcome this issue, in this work, we attempt to incorporate category information into the domain adaptation process by proposing Memory Guided Attention for Category-Aware Domain Adaptation (MeGA-CDA). The proposed method consists of employing category-wise discriminators to ensure category-aware feature alignment for learning domain-invariant discriminative features. However, since the category information is not available for the target samples, we propose to generate memory-guided category-specific attention maps which are then used to route the features appropriately to the corresponding category discriminator. The proposed method is evaluated on several benchmark datasets and is shown to outperform existing approaches.