DeconfuseTrack:Dealing with Confusion for Multi-Object Tracking

TL;DR

This work tackles confusion in multi-object tracking by introducing Decomposed Data Association (DDA), which breaks global data association into targeted sub-problems and selectively incorporates appearance cues, paired with Occlusion-aware NMS (ONMS) to preserve challenging detections. The resulting DeconfuseTrack framework demonstrates state-of-the-art performance on MOT17 and MOT20, significantly improving HOTA, IDF1, and AssA over strong baselines like ByteTrack, and shows good generalization to other trackers. The approach emphasizes interpretability and plug-and-play integration, avoiding learnable data-association modules while leveraging cues only when beneficial. Collectively, these contributions advance robust MOT under occlusion and dense scenes with a lean, modular design that reduces assignment confusion in practical settings.

Abstract

Accurate data association is crucial in reducing confusion, such as ID switches and assignment errors, in multi-object tracking (MOT). However, existing advanced methods often overlook the diversity among trajectories and the ambiguity and conflicts present in motion and appearance cues, leading to confusion among detections, trajectories, and associations when performing simple global data association. To address this issue, we propose a simple, versatile, and highly interpretable data association approach called Decomposed Data Association (DDA). DDA decomposes the traditional association problem into multiple sub-problems using a series of non-learning-based modules and selectively addresses the confusion in each sub-problem by incorporating targeted exploitation of new cues. Additionally, we introduce Occlusion-aware Non-Maximum Suppression (ONMS) to retain more occluded detections, thereby increasing opportunities for association with trajectories and indirectly reducing the confusion caused by missed detections. Finally, based on DDA and ONMS, we design a powerful multi-object tracker named DeconfuseTrack, specifically focused on resolving confusion in MOT. Extensive experiments conducted on the MOT17 and MOT20 datasets demonstrate that our proposed DDA and ONMS significantly enhance the performance of several popular trackers. Moreover, DeconfuseTrack achieves state-of-the-art performance on the MOT17 and MOT20 test sets, significantly outperforms the baseline tracker ByteTrack in metrics such as HOTA, IDF1, AssA. This validates that our tracking design effectively reduces confusion caused by simple global association.

Figures (5)

• Figure 1: Comparing different data association methods. (a) Global Association. (b) Decomposed Data Association (Ours)
• Figure 2: The overall pipeline of DeconfuseTrack. (1) Utilizing a detector to obtain the detection results for the current frame. (2) Employing ONMS to separate the detection results into reliable and unreliable detections. (3) Performing the first association using DDA. (4) Conducting the second association using unreliable detections and unassociated trajectories.
• Figure 3: Comparing the post-processing approaches of different trackers: (a) Ordinary trackers use NMS and discard low-scoring detections. (b) ByteTrack also utilizes NMS but retains low-scoring detections. (c) Our method employs ONMS and retains occluded detectons.
• Figure 4: Comparison of the performances of DeconfuseTrack under different detection confusion reduction factor. The results are from the validation set of MOT17.
• Figure 5: Visualization on the MOT17 validation set.