Research map / Video / Action recognition
Bridge-Prompt: Towards Ordinal Action Understanding 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 |
|---|---|---|---|
| use Bridge-Prompt-pretrained ViT features with ASFormer instead of conventional segmentation features | ASFormer · GTEA | F1@10: 90.9 -> 94.1 | Helped |
| use Bridge-Prompt-pretrained ViT features with ASFormer | ASFormer · 50Salads | F1@10: 86.6 -> 89.2 | Helped |
| fine-tune Bridge-Prompt features for activity recognition instead of fine-tuned I3D | Breakfast activity recognition · Breakfast | Acc(%): 74.83 -> 80.00 | Helped |
| replace early-fused ordinal prompts and learnable count token with vision-only fusion (ablation) | Bridge-Prompt · GTEA split #1 | F1@10: 91.0 -> 90.3 | Hurt |
| fuse ordinal prompts as positional embeddings by repeating the mapped vector (ablation) | Bridge-Prompt · GTEA split #1 | F1@10: 91.0 -> 89.1 | Hurt |
| fuse ordinal prompts as positional embeddings using an outer product (ablation) | Bridge-Prompt · GTEA split #1 | F1@10: 91.0 -> 88.7 | Hurt |
| use ordinal-prompt-weighted average pooling instead of early fusion (ablation) | Bridge-Prompt · GTEA split #1 | F1@10: 91.0 -> 91.8 | Helped |
| add early-fused ordinal prompts and a learnable count token rather than weighted-average fusion (ablation) | Bridge-Prompt · GTEA split #1 | F1@50: 79.1 -> 82.1 | Helped |
| add integrated-semantics loss (ablation) | Bridge-Prompt · GTEA split #1 | F1@10: 87.4 -> 88.6 | Helped |
| add statistical loss to semantic and integrated-semantics losses (ablation) | Bridge-Prompt · GTEA split #1 | F1@10: 88.6 -> 91.0 | Helped |
| transfer prompt-based action inference from making coffee to making tea | Bridge-Prompt · GTEA | top-1 Acc(%): 38.8 | Mixed |
Closest papers by meaning
- FineGym: A Hierarchical Video Dataset for Fine-Grained Action Understanding (CVPR 2020 · similarity 0.88)
- Intra- and Inter-Action Understanding via Temporal Action Parsing (CVPR 2020 · similarity 0.87)
- ActionAtlas: A VideoQA Benchmark for Domain-specialized Action Recognition (NeurIPS 2024 · similarity 0.87)
- Action Sets: Weakly Supervised Action Segmentation Without Ordering Constraints (CVPR 2018 · similarity 0.87)
- Kronecker Mask and Interpretive Prompts are Language-Action Video Learners (ICLR 2025 · similarity 0.86)
- Unsupervised Action Segmentation via Fast Learning of Semantically Consistent Actoms (AAAI 2024 · 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.