Research map / Video / Video-language understanding
Look Before You Speak: Visually Contextualized Utterances
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 |
|---|---|---|---|
| Add visual input to the text-only model | Future utterance prediction · HowToFUP | R@1: 60.90 -> 68.34 | Helped |
| Use a two-stream CoMVT rather than the single-stream multimodal baseline | Future utterance prediction · HowToFUP | R@1: 62.73 -> 63.35 | Helped |
| Add BERT pretraining (ablation) | CoMVT with scene features · HowToFUP | R@1: 63.35 -> 65.64 | Helped |
| Add the MLM loss (ablation) | CoMVT with scene features and BERT pretraining · HowToFUP | R@1: 65.64 -> 67.13 | Helped |
| Add combined object and scene features (ablation) | CoMVT with BERT pretraining and MLM · HowToFUP | R@1: 67.13 -> 67.79 | Helped |
| Increase CoTRM blocks from S=2 to S=4 with combined features and MLM (ablation) | CoMVT · HowToFUP | R@1: 67.79 -> 68.34 | Helped |
| Increase CoTRM blocks from S=2 to S=4 without MLM or object features (ablation) | CoMVT with scene features only · HowToFUP | R@1: 65.64 -> 61.05 | Hurt |
| Add compact feature set extraction (ablation) | CoMVT, S=2 · HowToFUP | R@1: 67.83 -> 67.79 | No clear effect |
| Add compact feature set extraction (ablation) | CoMVT, S=4 · HowToFUP | Gflops: 12.7 -> 7.7 (-39.3%) | Helped |
| Pretrain CoMVT on HowToFUP (ablation) | Future utterance prediction model · COIN-FUP | R@1: 39.11 -> 70.92 | Helped |
| Pretrain CoMVT on HowToFUP (ablation) | Next step prediction · COIN-NSP | R@1: 33.33 -> 37.46 | Helped |
| Pretrain CoMVT on HowToFUP (ablation) | Next step prediction · CrossTask-NSP | R@1: 33.73 -> 42.12 | Helped |
Closest papers by meaning
- Structure-Aware Multimodal Sequential Learning for Visual Dialog (AAAI 2024 · similarity 0.89)
- V^2Dial: Unification of Video and Visual Dialog via Multimodal Experts (CVPR 2025 · similarity 0.88)
- SayNext-Bench: Why Do LLMs Struggle with Next-Utterance Anticipation? (NeurIPS 2026 · similarity 0.88)
- Vision-Speech Models: Teaching Speech Models to Converse about Images (CVPR 2026 · similarity 0.87)
- Prismatic VLMs: Investigating the Design Space of Visually-Conditioned Language Models (ICML 2024 · similarity 0.87)
- TalkCuts: A Large-Scale Dataset for Multi-Shot Human Speech Video Generation (NeurIPS 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.