Research map / Video / Video-language understanding
Learning to Localize Actions in Instructional Videos with LLM-Based Multi-Pathway Text-Video Alignment
Yuxiao Chen, Kai Li, Wentao Bao, Deep Patel, Yu Kong, Martin Renqiang Min, Dimitris N. Metaxas
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 |
|---|---|---|---|
| train on LLM-steps instead of wikiHow-steps plus narrations (ablation) | MPTVA training · HT-Step | R@1: 35.9 -> 41.9 | Improved |
| add narrations to LLM-steps training (ablation) | MPTVA training · HT-Step | R@1: 41.9 -> 43.3 | Improved |
| use MPTVA pseudo-labels instead of timestamp labels for narrations (ablation) | narration training · HT-Step | R@1: 30.3 -> 32.1 | Improved |
| use narrations instead of wikiHow-steps with the same MPTVA supervision (ablation) | MPTVA training · HT-Step | R@1: 31.2 -> 32.1 | Improved |
| fuse all three alignment pathways rather than best single pathway (S-V-Short) (ablation) | MPTVA · HT-Step | R@1: 37.7 -> 41.9 | Improved |
| fuse all three pathways rather than best single pathway on CrossTask (ablation) | MPTVA · CrossTask | Avg. R@1: 41.6 -> 47.0 | Improved |
| add S-V-Short pathway to S-V-Long + S-N-V (ablation) | MPTVA · HT-Step | R@1: 37.9 -> 41.9 | Improved |
| train on S-N-V timestamp-derived labels instead of direct video alignment labels (ablation) | MPTVA · HT-Step | R@1: 34.2 vs 37.7 (S-V-Short) | Hurt |
| raise pseudo-label filter threshold from 0.65 to 0.70 (ablation) | MPTVA pseudo-labels · HT-Step | R@1: 40.6 -> 35.4 | Hurt |
| lower filter threshold from 0.65 to 0.60 (ablation) | MPTVA pseudo-labels · HT-Step | R@1: 40.6 -> 39.2 | Hurt |
| increase pseudo-label window size from 2 to 5 (ablation) | MPTVA pseudo-labels · HT-Step | R@1: 41.9 -> 40.6 | Hurt |
| decrease window size from 2 to 1 (ablation) | MPTVA pseudo-labels · HT-Step | R@1: 41.9 -> 40.3 | Hurt |
Closest papers by meaning
- VideoComp: Advancing Fine-Grained Compositional and Temporal Alignment in Video-Text Models (CVPR 2025 · similarity 0.88)
- Collaborative Weakly Supervised Video Correlation Learning for Procedure-Aware Instructional Video Analysis (AAAI 2024 · similarity 0.88)
- TOPA: Extending Large Language Models for Video Understanding via Text-Only Pre-Alignment (NeurIPS 2024 · similarity 0.87)
- SCHEMA: State CHangEs MAtter for Procedure Planning in Instructional Videos (ICLR 2024 · similarity 0.87)
- SF2T: Self-supervised Fragment Finetuning of Video-LLMs for Fine-Grained Understanding (CVPR 2025 · similarity 0.87)
- Masked Temporal Interpolation Diffusion for Procedure Planning in Instructional Videos (ICLR 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.
