WaspMOT 基于长时序多目标跟踪的 Trichogramma 蜂基准数据集

Jul 16, 2026

Upload by Ophelia

Multi-object tracking (MOT) benchmarks usually rely on short video clips, which overemphasizes short-term association and weakens assessment of long-term identity preservation. WaspMOT addresses this gap with controlled ecological experiments tracking Trichogramma wasps over long durations. The dataset includes 10 sequences of about 12,000 frames each, using dense MOTChallenge annotations and oracle detections. A closed-set protocol tests identity assignment across thousands of frames despite jumps, occlusions, and high visual similarity. Five tracking-by-detection methods are evaluated under a unified procedure, revealing consistent trajectory fragmentation and enabling baseline stitching.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages6

Reading Time15 min

Abstract

I. Introduction

Multi-object tracking and benchmark limitation

WaspMOT dataset design and closed-set scenario

Key challenges: jumps, occlusions, visual similarity

Oracle detections and unified evaluation protocol

Benchmark methods and trajectory fragmentation

Tracklet stitching baseline and implications

WaspMOT: A Benchmark for Long-Term Multi-Object Tracking of Trichogramma Wasps Tomasz Stanczyk1 ,3 ,∗ Yuan Gao2 ,3 ,∗ Hardik Agarwal 1 ,4 Seongroo Yoon 1 ,3 Tiantao Zhang5 Vincent Calcagno2 Francois Bremond 1 1Inria, Sophia Antipolis, France 2INRAE Institut Sophia Agrobiotech, Sophia Antipolis, France 3Universit Cte d’Azur, Sophia Antipolis, France 4Indian Institute of Technology Delhi, Delhi, India 5Institute of Plant Protection, Chinese Academy of Agricultural Sciences, Beijing, China ∗Joint first authors. arXiv :2607 .08729v 1 [ cs .CV] 9 Jul 2026 Abstract—Multi-object tracking (MOT) has achieved strong performance on benchmarks dominated by short video sequences. However, such datasets do not adequately evaluate long-term identity preservation, where objects must be tracked consistently over extended durations. We introduce WaspMOT, a benchmark designed to address this gap through long-duration tracking of Trichogramma wasps in controlled ecological experiments. The dataset contains 10 sequences of approximately 12,000 frames each (over 8 minutes at 25 FPS), with dense MOTChallenge annotations and oracle detections to isolate association performance. Unlike existing benchmarks, WaspMOT forms a closed-set tracking scenario where all individuals remain present throughout the sequence, requiring consistent identity assignment across thousands of frames despite abrupt jumps, occlusions, and highly similar appearance. We establish a benchmark by evaluating five tracking-by-detection methods, including ByteTrack, BoTSORT, C-BIoU, OC-SORT, and McByte, under a unified protocol. Results show that all methods suffer from significant trajectory fragmentation, highlighting the difficulty of long-term identity preservation even with perfect detections. A simple spatial tracklet stitching baseline consistently improves performance, indicating that substantial gains remain possible. WaspMOT provides a new benchmark for studying long-term association and reveals limitations of current tracking approaches that are not observable on conventional datasets. The benchmark will be made publicly available at the project repository: [https://github.com/tstanczyk95/WaspMOT/](https://github.com/tstanczyk95/WaspMOT/). Index Terms—multi-object tracking, long-term tracking, ecological surveillance, benchmark dataset I. INTRODUCTION Multi-object tracking (MOT) is a fundamental computer vision problem with applications in video surveillance, autonomous driving, robotics, and sports analytics. Its objective is to detect objects and maintain consistent identities overtime. Recent progress has been driven by benchmark datasets focusing on human-centered scenarios, including pedestrian tracking [1], [2], sports tracking [3], [4], and human motion analysis [5] . These datasets have enabled robust tracking algorithms capable of handling occlusions, appearance changes, and moderate motion. Fig. 1. Sample frame of a video with individuals inside an experimental arena. The image pixel dimensions are indicated in black and the real arena dimensions are indicated in red. However, existing MOT benchmarks are dominated by short sequences, typically lasting tens of seconds, where trajectories span only a limited portion of the video. As a result, trackers are primarily evaluated on short-term association rather than long-term identity preservation. This limits the ability to study scenarios where identities must be maintained consistently over extended durations. In ecological and biological monitoring, such long-term identity consistency is essential for analyzing individual behaviors and interactions. To address this gap, we introduce WaspMOT, a dataset designed for long-duration multi-object tracking of Trichogramma wasps in laboratory-controlled experiments. Each sequence contains approximately 12,000 frames (over 8 minutes at 25 FPS), and all individuals remain present throughout the entire video, forming a closed-set tracking

FAQs

WaspMOT解决了现有MOT基准在哪个关键评估缺口？

现有基准多基于短视频，主要评估短期关联；WaspMOT专注于长时长的身份保持（long-term identity preservation）。

WaspMOT数据集包含哪些实验设置与数据规模？

数据集包含10个序列，每个序列约12,000帧（25 FPS下超过8分钟），并采用MOTChallenge格式的稠密标注。

为什么WaspMOT的闭集跟踪场景会更难？

所有个体在整个序列中始终存在，但仍需在成千上万帧中保持一致身份，同时面临突发跳跃、遮挡以及高度相似的外观带来的关联困难。

文中如何保证评估更聚焦于关联性能而不是检测误差？

提供基于地面真值标注的oracle detections，隔离跟踪性能与检测错误，使评估聚焦于长期关联能力。

Jul 16, 2026

Upload by Ophelia