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

Researchers introduced TIDE, a dual-branch multi-object tracking system for dense schooling fish.

TIDE relies on geometric and structural cues, using Adaptive Geometric Correspondence IoU instead of heavy re-identification features.

The method is lightweight and scalable, aiming to support both edge deployment and higher-accuracy cloud setups.

It delivers competitive results on the MFT-Edge benchmark, and the authors released the dataset and code.