Research map / Video / Video generation
Point Prompting: Counterfactual Tracking with Video Diffusion Models
Ayush Shrivastava, Sanyam Mehta, Daniel Geng, Andrew Owens
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 |
|---|---|---|---|
| use the unedited first frame as a negative prompt (counterfactual enhancement guidance) (ablation) | Point Prompting · TAP-Vid DAVIS | tracking | Improved |
| add inpainting-based coarse-to-fine refinement (ablation) | Point Prompting · TAP-Vid DAVIS | positional accuracy | Improved |
| add color rebalancing to suppress red hues (ablation) | Point Prompting · TAP-Vid DAVIS | tracking | Improved |
| add local search window to tracker (ablation) | Point Prompting tracker · TAP-Vid DAVIS | AJ: 35.80 -> 38.92 | Improved |
| average positions of nearby red pixels (ablation) | Point Prompting tracker · TAP-Vid DAVIS | AJ: 39.08 -> 42.70 | Improved |
| use LAB instead of HSV color space (ablation) | Point Prompting tracker · TAP-Vid DAVIS | AJ: 42.70 -> 42.30 | Hurt |
| scale video model from Wan2.1-1.3B to 14B (ablation) | Point Prompting · TAP-Vid DAVIS | AJ: 44.58 -> 48.60 | Improved |
| upsample low-resolution inputs before generation (ablation) | Point Prompting · TAP-Vid Kubric | AJ: 31.51 -> 33.55 | Improved |
| run on original high-resolution DAVIS frames (ablation) | Point Prompting · TAP-Vid DAVIS | AJ: 42.21 -> 48.6 | Improved |
| set marker color to red versus other colors (ablation) | Point Prompting · TAP-Vid DAVIS subset | tracking | No clear effect |
| use guidance weight lambda = 8 (ablation) | Point Prompting · TAP-Vid DAVIS subset | tracking | Improved |
| distill generator tracks into CoTracker | CoTracker trained from scratch · TAP-Vid DAVIS | AJ: teacher 38.78 vs distilled 37.17 | Improved |
Closest papers by meaning
- Emergent Temporal Correspondences from Video Diffusion Transformers (NeurIPS 2025 · similarity 0.90)
- Track4Gen: Teaching Video Diffusion Models to Track Points Improves Video Generation (CVPR 2025 · similarity 0.90)
- Video Diffusion Models Excel at Tracking Similar-Looking Objects Without Supervision (NeurIPS 2025 · similarity 0.89)
- DiffusionTrack: Point Set Diffusion Model for Visual Object Tracking (CVPR 2024 · similarity 0.88)
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.88)
- Training-free Camera Control for Video Generation (ICLR 2025 · similarity 0.88)
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.
