Research map / Video / Action recognition
Action Modifiers: Learning From Adverbs in Instructional Videos
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Learn shared linear action-modifier transformations and action-query attention | Action Modifiers · Video-to-adverb, All | mAP: 0.621 -> 0.719 | Helped |
| Learn shared linear action-modifier transformations and action-query attention | Action Modifiers · Video-to-adverb, Antonym | P@1: 0.728 -> 0.808 | Helped |
| Use the proposed method for adverb-to-video retrieval | Action Modifiers · Adverb-to-video, Antonym | mAP: 0.618 -> 0.657 | Helped |
| Use the proposed method for unrestricted adverb-to-video retrieval | Action Modifiers · Adverb-to-video, All | mAP: 0.350 -> 0.329 | Hurt |
| Use a learned linear modifier instead of fixed or learned translations or a nonlinear transform (ablation) | Action Modifiers · Video-to-adverb, Antonym | P@1: 0.735, 0.749, 0.742 -> 0.808 | Helped |
| Add an initial action-only training stage before learning adverbs and modifiers (ablation) | Action Modifiers · Video-to-adverb, Antonym | P@1: 0.774 -> 0.808 | Helped |
| Use action-query multi-head scaled dot-product attention instead of single-clip, average, class-agnostic, or class-specific aggregation (ablation) | Action Modifiers · Video-to-adverb, Antonym | P@1: 0.705, 0.716, 0.708, 0.728 -> 0.808 | Helped |
| Use action-query multi-head scaled dot-product attention (ablation) | Action Modifiers · Video-to-action | mAP: 0.246 -> 0.692 | Helped |
| Use two separate action and adverb losses instead of a single loss with mixed negatives (ablation) | Action Modifiers · Video-to-adverb, Antonym | P@1: +0.03 P@1 | Helped |
| Use only the antonym as the adverb-loss negative instead of another adverb (ablation) | Action Modifiers · Video-to-adverb, Antonym | P@1: +0.04 P@1 | Helped |
| Jointly learn action-relevant attention and action modifiers instead of using W-TALC localization followed by averaging (ablation) | Action Modifiers · Video-to-adverb, Antonym | P@1: 0.739 -> 0.808 | Helped |
| Use the action embedding as the attention query instead of a one-hot action, adverb, or action-adverb representation (ablation) | Action Modifiers · Video-to-adverb, Antonym | P@1: 0.736, 0.702, 0.731, 0.728 -> 0.774 | Helped |
Closest papers by meaning
- Learning Action Changes by Measuring Verb-Adverb Textual Relationships (CVPR 2023 · similarity 0.94)
- How Do You Do It? Fine-Grained Action Understanding With Pseudo-Adverbs (CVPR 2022 · similarity 0.93)
- Further Understanding Videos through Adverbs: A New Video Task (AAAI 2020 · similarity 0.89)
- Semi-Weakly-Supervised Learning of Complex Actions From Instructional Task Videos (CVPR 2022 · similarity 0.87)
- Detours for Navigating Instructional Videos (CVPR 2024 · similarity 0.86)
- Kronecker Mask and Interpretive Prompts are Language-Action Video Learners (ICLR 2025 · similarity 0.86)
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.