SFDATrack: Generalized Source-Free Domain Adaptive Tracking Under Adverse Weather Conditions

2026-06-30

AI Summary

This work addresses the challenge of visual object tracking under adverse weather conditions, where existing methods rely heavily on large-scale data from both source and target domains and thus fail in practical scenarios where source-domain data are unavailable. To overcome this limitation, we propose SFDATrack, the first framework enabling universal domain-adaptive tracking without access to source-domain data. SFDATrack leverages a mean-teacher architecture combined with a Dual Interactive Mamba module to extract weather-robust features, and introduces a Hyperspherical Prototype Projection (HPP) module that jointly models domain-invariant and domain-specific prototypes within an implicit hyperspherical space. Integrated with feature distillation and tailored data augmentation, SFDATrack significantly outperforms state-of-the-art methods across multiple adverse-weather benchmarks, demonstrating exceptional generalization capability and tracking robustness.

Abstract

Domain adaptive visual object tracking under adverse weather conditions has garnered significant attention in recent years. Despite the impressive performance, existing methods heavily rely on the large-scale video frames from both source and target domains, which is impractical under rigid resource constraints where source data is unavailable. To overcome this limitation, we propose SFDATrack, a generalized source-free domain adaptive tracker that merely leverages adverse weather samples from the target domain for robust state estimation. Specifically, SFDATrack first employs a mean-teacher backbone with Dual Interactive Mamba (DIM) blocks to distill the candidate target tokens that are resilient to weather variations from classified, augmented samples. Afterwards, we introduce a hyperspherical prototype projection (HPP) module to project these tokens onto multi-domain prototypes within a latent hyperspherical space. By enforcing both domain-specific and domain-invariant properties of the multi-domain prototypes, SFDATrack can be seamlessly adapted to diverse weather conditions with powerful generalizability. Extensive experiments evaluated on various benchmarks demonstrate that SFDATrack achieves superior performance compared to state-of-the-art approaches. The code is available at https://github.com/watcherBR0/sfdatrack.

Problem

• source-free domain adaptation
• visual object tracking
• adverse weather conditions
• domain generalization

Innovation

• Source-Free Domain Adaptation
• Visual Object Tracking
• Adverse Weather Conditions
• Hyperspherical Prototype Projection
• Dual Interactive Mamba