Research map / Detection and segmentation / Tracking
Online Dense Point Tracking with Streaming Memory
Qiaole Dong, Yanwei Fu
AI-extracted. The measured effects below were read by AI from the paper's full text (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 memory bank and memory reading (ablation) | SPOT · CVO (Extended, 10-frame training) | EPE (all): 38.98 -> 6.42 | Improved |
| add sensory memory (ablation) | SPOT · CVO (Extended, 10-frame training) | EPE (all): 8.64 -> 6.42 | Improved |
| add query projector for memory reading (ablation) | SPOT · CVO (Extended, 10-frame training) | EPE (all): 6.48 -> 6.42 | Improved |
| add feature fusion layer (ablation) | SPOT · CVO (Extended, 10-frame training) | EPE (all): NaN -> 6.42 | Improved |
| use linear splatting instead of average splatting (ablation) | SPOT · CVO (Extended, 10-frame training) | EPE (all): 7.34 -> 6.42 | Improved |
| use linear splatting instead of softmax splatting (ablation) | SPOT · CVO (Extended, 10-frame training) | EPE (all): 7.04 -> 6.42 | Improved |
| use linear splatting instead of summation splatting (ablation) | SPOT · CVO (Extended, 10-frame training) | EPE (all): 7.17 -> 6.42 | Improved |
| add hidden-state warm-start (ablation) | SPOT · CVO (Extended, 10-frame training) | EPE (all): 7.94 -> 6.42 | Improved |
| add flow warm-start (ablation) | SPOT · CVO (Extended, 10-frame training) | EPE (all): 6.49 -> 6.42 | Improved |
| increase training video length from 7 to 24 frames (ablation) | SPOT · CVO (Extended) | EPE (all): 7.14 -> 4.77 | Improved |
| increase memory length from 1 to 3 (ablation) | SPOT · CVO (Extended) | EPE (all): 5.00 -> 4.77 | Improved |
| increase memory length from 3 to 6 (ablation) | SPOT · CVO (Extended) | EPE (all): 4.77 -> 5.06 | Hurt |
Closest papers by meaning
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.91)
- Dense Optical Tracking: Connecting the Dots (CVPR 2024 · similarity 0.90)
- FlowTrack: Revisiting Optical Flow for Long-Range Dense Tracking (CVPR 2024 · similarity 0.89)
- TrackIME: Enhanced Video Point Tracking via Instance Motion Estimation (NeurIPS 2024 · similarity 0.89)
- TrackCraft3R: Repurposing Video Diffusion Transformers for Dense 3D Tracking (NeurIPS 2026 · similarity 0.88)
- Is This Tracker On? A Benchmark Protocol for Dynamic Tracking (NeurIPS 2025 · similarity 0.87)
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.
