DA-DETR: Domain Adaptive Detection Transformer with Information Fusion

TL;DR

DA-DETR introduces a novel domain adaptive detection transformer that fuses CNN and Transformer features for effective domain transfer, achieving superior performance across multiple benchmarks.

Contribution

It proposes a new CNN-Transformer Blender (CTBlender) for feature fusion, enhancing domain adaptation in detection transformers.

Findings

DA-DETR outperforms existing methods on domain adaptation benchmarks.

The CTBlender effectively fuses semantic and spatial features across domains.

The approach simplifies domain adaptation with a transformer-based architecture.

Abstract

The recent detection transformer (DETR) simplifies the object detection pipeline by removing hand-crafted designs and hyperparameters as employed in conventional two-stage object detectors. However, how to leverage the simple yet effective DETR architecture in domain adaptive object detection is largely neglected. Inspired by the unique DETR attention mechanisms, we design DA-DETR, a domain adaptive object detection transformer that introduces information fusion for effective transfer from a labeled source domain to an unlabeled target domain. DA-DETR introduces a novel CNN-Transformer Blender (CTBlender) that fuses the CNN features and Transformer features ingeniously for effective feature alignment and knowledge transfer across domains. Specifically, CTBlender employs the Transformer features to modulate the CNN features across multiple scales where the high-level semantic information and the low-level spatial information are fused for accurate object identification and localization. Extensive experiments show that DA-DETR achieves superior detection performance consistently across multiple widely adopted domain adaptation benchmarks.