Research map / Detection and segmentation / Tracking
Prototypical Cross-Attention Networks for Multiple Object Tracking and Segmentation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add long-term temporal memory (length 32) instead of using a single frame (ablation) | PCAN · YouTube-VIS validation | AP: 32.5 -> 35.4 | Helped |
| Aggregate temporal features across FPN layers P3-P5 rather than using P5 only (ablation) | PCAN · YouTube-VIS validation | AP: 32.9 -> 33.9 | Helped |
| Use prototypical cross-attention instead of non-local attention | YouTube-VIS model, length 8 · YouTube-VIS validation | AP: 33.6 -> 34.2 | Helped |
| Add 30 foreground and 30 background instance prototypes rather than none (ablation) | PCAN · YouTube-VIS validation | AP: 32.5 -> 33.9 | Helped |
| Use one foreground and one background prototype rather than no instance prototypes (ablation) | PCAN · YouTube-VIS validation | AP: 32.5 -> 32.7 | No clear effect |
| Recurrently propagate and update instance prototypes with momentum 0.2 rather than not propagating them (ablation) | PCAN · YouTube-VIS validation | AP: 33.5 -> 34.3 | Helped |
| Use 128 rather than 64 frame-level prototypes (ablation) | PCAN · YouTube-VIS validation | AP: 34.2 -> 34.1 | No clear effect |
| Use 6 rather than 8 EM iterations (ablation) | PCAN · YouTube-VIS validation | AP: 33.6 -> 33.9 | Helped |
| Add frame-level PCAM (ablation) | PCAN · KITTI-MOTS validation, Car | Car-MOTSA: 87.3 -> 89.6 | Helped |
| Add instance-level PCAM (ablation) | PCAN · KITTI-MOTS validation, Pedestrian | Ped-MOTSA: 65.8 -> 66.4 | Helped |
| Use PCAN rather than SG-Net with ResNet-50 | YouTube-VIS validation · YouTube-VIS validation | AP: 34.8 -> 36.1 | Helped |
| Use PCAN rather than QDTrack-mots-fix | BDD100K segmentation tracking validation · BDD100K segmentation tracking validation | mMOTSA: 23.5 -> 27.4 | Helped |
Closest papers by meaning
- Target-Aware Object Discovery and Association for Unsupervised Video Multi-Object Segmentation (CVPR 2021 · similarity 0.89)
- Learning to Track Instances without Video Annotations (CVPR 2021 · similarity 0.89)
- See More, Know More: Unsupervised Video Object Segmentation With Co-Attention Siamese Networks (CVPR 2019 · similarity 0.88)
- LiVOS: Light Video Object Segmentation with Gated Linear Matching (CVPR 2025 · similarity 0.87)
- A Distractor-Aware Memory for Visual Object Tracking with SAM2 (CVPR 2025 · similarity 0.86)
- Matching Anything by Segmenting Anything (CVPR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.