ABRA: Teleporting Fine-Tuned Knowledge Across Domains for Open-Vocabulary Object Detection

TL;DR

ABRA is a novel method that transfers class-specific detection knowledge across domains without target domain training data, effectively addressing domain shifts in open-vocabulary object detection.

Contribution

ABRA introduces a geometric transport approach in weight space to adapt pretrained detectors to new domains without access to target domain images.

Findings

Effective transfer of class knowledge across domains

Improved detection performance under adverse conditions

Applicable without target domain training data

Abstract

Although recent Open-Vocabulary Object Detection architectures, such as Grounding DINO, demonstrate strong zero-shot capabilities, their performance degrades significantly under domain shifts. Moreover, many domains of practical interest, such as nighttime or foggy scenes, lack large annotated datasets, preventing direct fine-tuning. In this paper, we introduce Aligned Basis Relocation for Adaptation(ABRA), a method that transfers class-specific detection knowledge from a labeled source domain to a target domain where no training images containing these classes are accessible. ABRA formulates this adaptation as a geometric transport problem in the weight space of a pretrained detector, aligning source and target domain experts to transport class-specific knowledge. Extensive experiments across challenging domain shifts demonstrate that ABRA successfully teleports class-level specialization under multiple adverse conditions. Our code will be made public upon acceptance.