Research map / Video / Video generation
Self-Paced and Self-Corrective Masked Prediction for Movie Trailer Generation
Sidan Zhu, Hongteng Xu, Dixin Luo
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 self-paced mask ratio instead of random (ablation) | SSMP · Test-ALL | F1: 0.1377 -> 0.1996 | Improved |
| use self-paced instead of linearly increased mask ratio (ablation) | SSMP · Test-ALL | F1: 0.1915 -> 0.1996 | Improved |
| increase mask ratio over training rather than decrease (ablation) | SSMP · Test-ALL | F1: 0.1760 -> 0.1915 | Improved |
| add self-correction (re-masking) instead of greedy fill (ablation) | SSMP · Test-ALL | F1: 0.1958 -> 0.1996 | Improved |
| add self-correction instead of greedy under relaxed metric (ablation) | SSMP · Test-74 (R=2) | F1: 0.544 -> 0.622 | Improved |
| use cross-entropy instead of MSE loss (ablation) | SSMP · Test-ALL | F1: 0.1211 -> 0.1996 | Improved |
| set mask ratio momentum mu_t to 0.1 rather than 0.9 (ablation) | SSMP · Test-ALL | F1: 0.1744 -> 0.1996 | Improved |
| set accuracy momentum mu_a to 0.98 rather than 1.0 (ablation) | SSMP · Test-ALL | F1: 0.1780 -> 0.1996 | Improved |
| add monotonicity constraint on mask ratio (ablation) | SSMP · Test-ALL | mask ratio: around 0.9 | Improved |
| replace ImageBind with DINOv3 (ablation) | SSMP · Test-ALL | F1: 0.1996 -> 0.1990 | No clear effect |
| replace ImageBind with SigLIP2 (ablation) | SSMP · Test-ALL | F1: 0.1996 -> 0.1936 | No clear effect |
| go from 5 layers to 8 layers (ablation) | SSMP · Test-ALL | F1: 0.1710 -> 0.1694 | No clear effect |
Closest papers by meaning
- Towards Automated Movie Trailer Generation (CVPR 2024 · similarity 0.88)
- MaskINT: Video Editing via Interpolative Non-autoregressive Masked Transformers (CVPR 2024 · similarity 0.83)
- SG-I2V: Self-Guided Trajectory Control in Image-to-Video Generation (ICLR 2025 · similarity 0.83)
- Boximator: Generating Rich and Controllable Motions for Video Synthesis (ICML 2024 · similarity 0.83)
- Taming Teacher Forcing for Masked Autoregressive Video Generation (CVPR 2025 · similarity 0.82)
- Through-The-Mask: Mask-based Motion Trajectories for Image-to-Video Generation (CVPR 2025 · similarity 0.82)
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.
