Research map / Vision-language and multimodal / Cross-modal learning
Look at What I’m Doing: Self-Supervised Spatial Grounding of Narrations in Instructional Videos
Reuben Tan, Bryan A. Plummer, Kate Saenko, Hailin Jin, Bryan Russell
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 sentence-level rather than word-level contrastive loss (ablation) | CoMMA · YouCook2-Interactions | localization accuracy (%): 34.91 -> 55.80 | Improved |
| Use CoMMA sentence-level grounding instead of the strongest listed baseline, Shallow Attention | YouCook2-Interactions · YouCook2-Interactions | localization accuracy (%): 48.30 -> 55.80 | Improved |
| Self-train CoMMA on narrated video clips for object localization | CoMMA · YouCook2-BB validation | full localization (%): 46.95 -> 59.25 | Improved |
| Add a second cross-attention layer (ablation) | CoMMA · YouCook2-Interactions | localization accuracy (%): 48.30 -> 51.09 | Improved |
| Add full spatiotemporal self-attention instead of spatial-only self-attention (ablation) | CoMMA · YouCook2-Interactions | localization accuracy (%): 53.88 -> 55.80 | Improved |
| Use temporal-only self-attention instead of full spatiotemporal self-attention (ablation) | CoMMA · YouCook2-Interactions | localization accuracy (%): 52.72 -> 55.80 | Hurt |
| Use separate spatial and temporal self-attention instead of full spatiotemporal self-attention (ablation) | CoMMA · YouCook2-Interactions | localization accuracy (%): 52.15 -> 55.80 | Hurt |
| Add a second cross-attention layer to sentence-loss-only CoMMA (ablation) | CoMMA hybrid phrase-grounding model · Flickr30K | R@1: 35.43 -> 39.03 | Improved |
| Add self-attention to sentence-loss-only CoMMA (ablation) | CoMMA hybrid phrase-grounding model · Flickr30K | R@1: 39.03 -> 39.31 | Improved |
| Add sentence-level loss to the one-layer word-loss hybrid (ablation) | CoMMA hybrid phrase-grounding model · Flickr30K | R@1: 49.99 -> 52.43 | Improved |
| Add a second cross-attention layer to the word + sentence loss hybrid (ablation) | CoMMA hybrid phrase-grounding model · Flickr30K | R@1: 52.43 -> 53.80 | Improved |
| Add self-attention to the two-layer word + sentence loss hybrid (ablation) | CoMMA hybrid phrase-grounding model · Flickr30K | R@1: 53.80 -> 48.53 | Hurt |
Closest papers by meaning
- What When and Where? Self-Supervised Spatio-Temporal Grounding in Untrimmed Multi-Action Videos from Narrated Instructions (CVPR 2024 · similarity 0.90)
- Contextual Self-paced Learning for Weakly Supervised Spatio-Temporal Video Grounding (ICLR 2025 · similarity 0.88)
- SAMA: Towards Multi-Turn Referential Grounded Video Chat with Large Language Models (NeurIPS 2025 · similarity 0.87)
- VideoGEM: Training-free Action Grounding in Videos (CVPR 2025 · similarity 0.87)
- Improved Visual Grounding through Self-Consistent Explanations (CVPR 2024 · similarity 0.87)
- Learning to Segment Referred Objects from Narrated Egocentric Videos (CVPR 2024 · 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.
