When Fish Look Alike: Tracking Identities with Dual-branch Elasticity

2026-07-28

AI Summary

This work addresses the challenge of multi-object tracking for dense, homogeneous targets—such as fish schools—where appearance similarity, severe occlusion, and non-rigid deformations hinder performance, and state-of-the-art high-accuracy methods are impractical for edge deployment. To bridge the gap between edge efficiency and cloud-level accuracy, we propose TIDE, a novel dual-branch elastic framework featuring a lightweight L-branch for efficient edge inference and a scalable S-branch for high-precision cloud analysis. TIDE eliminates reliance on appearance features by leveraging adaptive geometric correspondence IoU and structural consistency modeling for robust association, while decoupling algorithm design from hardware constraints to enable seamless edge-cloud collaboration. On the MFT-Edge benchmark, the L-branch achieves 28.43 HOTA at only 20.47G FLOPs—38.7× fewer than SU-T—while the S-branch attains 29.98 HOTA, effectively reconciling computational efficiency with tracking accuracy.

Abstract

Tracking dense, homogeneous targets like schooling fish remains a major challenge for multiple object tracking due to extreme inter-individual homogeneity, severe physical clustering, and rapid non-rigid deformations. While heavy-backbone separated detection and embedding trackers like SU-T push accuracy boundaries using complex Re-Identification networks, their computational overhead prohibits edge deployment. Furthermore, these modules often fail when appearance features degrade under severe occlusions. To overcome this, we propose Tracking Identities with Dual-branch Elasticity (TIDE). Bypassing expensive appearance cues, TIDE utilizes the Adaptive Geometric Correspondence IoU, an association mechanism leveraging spatial and structural consistency to robustly handle complex morphological variations. Crucially, TIDE introduces system-level deployment elasticity, decoupling the algorithmic pipeline from strict hardware constraints. Evaluations on the MFT-Edge benchmark demonstrate that our Lightweight L-branch achieves a competitive HOTA of 28.43 using merely 20.47G FLOPs. This represents a 38.7-fold computational reduction compared to upper bounds like SU-T, directly facilitating real-time edge deployment. Concurrently, our Scalable S-branch establishes a 29.98 HOTA, successfully bridging the gap between high-precision cloud analysis and efficient edge tracking. The dataset and codes are released at https://vranlee.github.io/TIDE/.

Problem

• multiple object tracking
• homogeneous targets
• occlusion
• edge deployment
• appearance degradation

Innovation

• Dual-branch Elasticity
• Adaptive Geometric Correspondence IoU
• Edge Deployment
• Appearance-free Tracking
• Multiple Object Tracking