Research map / Video / Video segmentation and tracking
MoSiC: Optimal-Transport Motion Trajectory for Dense Self-Supervised Learning
Mohammadreza Salehi, Shashanka Venkataramanan, Ioana Simion, Efstratios Gavves, Cees G. M. Snoek, Yuki M Asano
Read the paperCheck your idea against 137,217 papers
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 EMA teacher (ablation) | MoSiC · Pascal VOC in-context (DINO ViT-S ablation) | mIoU: 50.5 -> 51.5 | Improved |
| apply loss only to visible points (ablation) | MoSiC · Pascal VOC in-context (DINO ViT-S ablation) | mIoU: 50.5 -> 51.5 | Improved |
| use 16x16 tracking grid instead of 8x8 (ablation) | MoSiC · Pascal VOC in-context (DINO ViT-S ablation) | mIoU: 49.2 -> 51.5 | Improved |
| increase tracking grid from 16x16 to 32x32 (ablation) | MoSiC · Pascal VOC in-context (DINO ViT-S ablation) | mIoU: 51.5 -> 50.5 | Hurt |
| use 100 prototypes instead of 50 (ablation) | MoSiC · Pascal VOC in-context (DINO ViT-S ablation) | mIoU: 46.5 -> 51.5 | Improved |
| increase prototypes from 100 to 300 (ablation) | MoSiC · Pascal VOC in-context (DINO ViT-S ablation) | mIoU: 51.5 -> 50.1 | No clear effect |
| add crop augmentation with scale [0.4,1] (ablation) | MoSiC · Pascal VOC in-context (DINO ViT-S ablation) | mIoU: 49.1 -> 51.5 | Improved |
| add 10% token masking (ablation) | MoSiC · Pascal VOC in-context (DINO ViT-S ablation) | mIoU: 51.1 -> 51.5 | Improved |
| increase masking ratio to 40% (ablation) | MoSiC · Pascal VOC in-context (DINO ViT-S ablation) | mIoU: 51.5 -> 49.9 | Hurt |
| use 12 frames instead of 2 (ablation) | MoSiC · Pascal VOC in-context (DINO ViT-S ablation) | mIoU: 39.3 -> 51.5 | Improved |
| lengthen clip from 0.5 s to 3.2 s (ablation) | MoSiC · Pascal VOC in-context (DINO ViT-S ablation) | mIoU: 47.6 -> 51.5 | Improved |
| use CoTracker3 instead of RAFT point tracker (ablation) | MoSiC · Pascal VOC in-context (DINOv2 ViT-S) | mIoU: 77.6 -> 78.2 | Improved |
Closest papers by meaning
- Learning Segmentation from Point Trajectories (NeurIPS 2024 · similarity 0.88)
- Video Diffusion Models Excel at Tracking Similar-Looking Objects Without Supervision (NeurIPS 2025 · similarity 0.87)
- Tracktention: Leveraging Point Tracking to Attend Videos Faster and Better (CVPR 2025 · similarity 0.87)
- TrackIME: Enhanced Video Point Tracking via Instance Motion Estimation (NeurIPS 2024 · similarity 0.87)
- Moving Off-the-Grid: Scene-Grounded Video Representations (NeurIPS 2024 · similarity 0.87)
- GMOS: Grounding Moving Object Segmentation in 3D Space and Time (NeurIPS 2026 · 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.
