Research on the Long-tailed Distribution of Data in Bee Tracking
The dataset and algorithms have always been critical points in multi-object tracking (MOT) tasks, as the long-tailed problem of the dataset directly impacts the results of tracking algorithms. To address this issue, we introduce the BeeTrack dataset, which highlights the long-tailed problem. The lon...
| id |
Zheng Y.; Idrus Z.; Xu W. |
|---|---|
| spelling |
Zheng Y.; Idrus Z.; Xu W. 2-s2.0-85219167939 Research on the Long-tailed Distribution of Data in Bee Tracking 2025 Proceedings of 2024 3rd International Conference on Artificial Intelligence and Intelligent Information Processing, AIIIP 2024 10.1145/3707292.3707350 https://www.scopus.com/inward/record.uri?eid=2-s2.0-85219167939&doi=10.1145%2f3707292.3707350&partnerID=40&md5=f4eb244799b2d63ae53debbcb1b44e81 The dataset and algorithms have always been critical points in multi-object tracking (MOT) tasks, as the long-tailed problem of the dataset directly impacts the results of tracking algorithms. To address this issue, we introduce the BeeTrack dataset, which highlights the long-tailed problem. The long-tailed issue primarily arises from the uneven distribution of data. To mitigate this, we incorporate a historical memory mechanism and linear interpolation in our approach. Specifically, we leverage the historical states of unmatched trajectories to infer their current states. Simultaneously, we use interpolation to complete the motion trajectories of the targets, enhancing the algorithm’s predictive capability regarding target behavior. The experimental results show that our improved network Higher Order Tracking Accuracy (HOTA) and Multiple Object Tracking Accuracy (MOTA) reach 46.2 and 62.8 respectively, with a reduction of 87 Identity Switches (IDS) and a significant overall performance improvement. Our method shows promise in effectively addressing the long-tailed problem, while also presenting the challenging BeeTrack dataset, offering new insights for the development of MOT. © 2024 Copyright held by the owner/author(s). Association for Computing Machinery, Inc English Conference paper |
| author |
2-s2.0-85219167939 |
| spellingShingle |
2-s2.0-85219167939 Research on the Long-tailed Distribution of Data in Bee Tracking |
| author_facet |
2-s2.0-85219167939 |
| author_sort |
2-s2.0-85219167939 |
| title |
Research on the Long-tailed Distribution of Data in Bee Tracking |
| title_short |
Research on the Long-tailed Distribution of Data in Bee Tracking |
| title_full |
Research on the Long-tailed Distribution of Data in Bee Tracking |
| title_fullStr |
Research on the Long-tailed Distribution of Data in Bee Tracking |
| title_full_unstemmed |
Research on the Long-tailed Distribution of Data in Bee Tracking |
| title_sort |
Research on the Long-tailed Distribution of Data in Bee Tracking |
| publishDate |
2025 |
| container_title |
Proceedings of 2024 3rd International Conference on Artificial Intelligence and Intelligent Information Processing, AIIIP 2024 |
| container_volume |
|
| container_issue |
|
| doi_str_mv |
10.1145/3707292.3707350 |
| url |
https://www.scopus.com/inward/record.uri?eid=2-s2.0-85219167939&doi=10.1145%2f3707292.3707350&partnerID=40&md5=f4eb244799b2d63ae53debbcb1b44e81 |
| description |
The dataset and algorithms have always been critical points in multi-object tracking (MOT) tasks, as the long-tailed problem of the dataset directly impacts the results of tracking algorithms. To address this issue, we introduce the BeeTrack dataset, which highlights the long-tailed problem. The long-tailed issue primarily arises from the uneven distribution of data. To mitigate this, we incorporate a historical memory mechanism and linear interpolation in our approach. Specifically, we leverage the historical states of unmatched trajectories to infer their current states. Simultaneously, we use interpolation to complete the motion trajectories of the targets, enhancing the algorithm’s predictive capability regarding target behavior. The experimental results show that our improved network Higher Order Tracking Accuracy (HOTA) and Multiple Object Tracking Accuracy (MOTA) reach 46.2 and 62.8 respectively, with a reduction of 87 Identity Switches (IDS) and a significant overall performance improvement. Our method shows promise in effectively addressing the long-tailed problem, while also presenting the challenging BeeTrack dataset, offering new insights for the development of MOT. © 2024 Copyright held by the owner/author(s). |
| publisher |
Association for Computing Machinery, Inc |
| issn |
|
| language |
English |
| format |
Conference paper |
| accesstype |
|
| record_format |
scopus |
| collection |
Scopus |
| _version_ |
1828987858409488384 |