LATrack: Limited Attention for Visual Object Tracking
The use of temporal information is becoming increasingly important in mainstream visual object trackers. Mainstream trackers typically interact trajectory information with image features. However, existing methods of interaction cannot effectively utilize trajectory information, and the significance...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10820357/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The use of temporal information is becoming increasingly important in mainstream visual object trackers. Mainstream trackers typically interact trajectory information with image features. However, existing methods of interaction cannot effectively utilize trajectory information, and the significance of the interaction remains unclear. To address these issues, we propose a Limited Attention module (LA module). The LA module more effectively utilizes image features by masking certain image features based on historical trajectory information or prediction information. Based on the LA module, we propose Limited Attention Track (LATrack), which can make more effective use of trajectory information. LATrack can continuously approach the target object by utilizing the predicted coordinates of the historical trajectory, thereby obtaining the object’s position in the current frame. Our model excels in handling challenges such as motion blur, distraction from similar objects, and occlusion. LATrack demonstrates excellent performance across multiple datasets, notably achieving 76.7% AO and 53.8% AUC on the GOT-10k and <inline-formula> <tex-math notation="LaTeX">$\mathrm {LaSOT_{ext}}$ </tex-math></inline-formula> datasets, respectively. |
|---|---|
| ISSN: | 2169-3536 |