Orthogonal Knowledge Refreshing for Domain-Incremental Object Detection

2026-07-19

AI Summary

This work addresses the challenges of catastrophic forgetting and inter-domain interference in domain-incremental object detection by proposing a replay-free, continual learning approach that preserves previously acquired knowledge. The method enables parameter-efficient adaptation through orthogonal, domain-specific low-rank subspaces and incorporates gradient orthogonal projection together with a topology-aware consistency mechanism to suppress interference while maintaining semantic structural stability during new task acquisition. Notably, it is the first to integrate gradient projection with semantic topology alignment for domain-incremental detection. Experimental results demonstrate significant improvements over the current state-of-the-art replay-free methods, achieving gains of 5.6% and 6.5% in mAP on Pascal VOC and BDD100K, respectively.

Abstract

Domain-incremental object detection (DIOD) requires models to continually adapt to new domains while preserving prior knowledge. Recently, parameter-efficient fine-tuning offers a promising avenue, wherein a pre-trained model is frozen and a small number of learnable parameters are injected for downstream tasks. However, these methods risk overwriting critical past knowledge, triggering inter-domain interference and performance degradation. To address this challenge, we propose Orthogonal Knowledge Refreshing (OKR), a simple yet effective framework for DIOD. OKR incrementally constructs independent domain-specific subspaces via dedicated low-rank branches for each domain, which are seamlessly fused for a holistic decision, enabling conflict-free capacity expansion without domain selection during inference. To minimize knowledge interference during fusion, we present a gradient-based orthogonal refreshing strategy that projects gradient updates of new domains onto the orthogonal complement of the fused historical subspace, supporting continual adaptation without forgetting. Moreover, to mitigate semantic fragmentation across domains, we enforce topology-aware consistency, aligning the semantic structures of old and new domains. Extensive experiments validate the superiority of OKR, outperforming the best exemplar-free method by significant margins of +5.6% and +6.5% mAP on the Pascal VOC and BDD100K series, respectively.

Problem

• domain-incremental object detection
• catastrophic forgetting
• knowledge interference
• continual learning
• parameter-efficient fine-tuning

Innovation

• Orthogonal Knowledge Refreshing
• Domain-Incremental Object Detection
• Low-Rank Subspace
• Gradient Orthogonalization
• Topology-Aware Consistency