Research map / Video / Video-language understanding
ER3: A Unified Framework for Event Retrieval, Recognition and Recounting
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 |
|---|---|---|---|
| Align frame features into a video imprint and aggregate its active counting-grid descriptors instead of averaging frame descriptors (ablation) | ER3 event retrieval · EVVE | mAP: 47.3 -> 52.3 | Helped |
| Align frame features into a video imprint and aggregate its active counting-grid descriptors instead of averaging frame descriptors (ablation) | ER3 event retrieval · EVVE+100K | mAP: 38.7 -> 42.9 | Helped |
| Add average query expansion to counting-grid aggregation | ER3 event retrieval · EVVE | mAP: 52.3 -> 58.5 | Helped |
| Add Difference of Neighborhood re-ranking to counting-grid aggregation | ER3 event retrieval · EVVE | mAP: 52.3 -> 60.1 | Helped |
| Add the reasoning network to counting-grid aggregation (ablation) | ER3 event recognition · CCV | mAP: 79.1 -> 79.9 | Helped |
| Add the reasoning network to counting-grid aggregation (ablation) | ER3 event recognition · MED14 | mAP: 33.7 -> 36.9 | Helped |
| Fuse audio and motion features with the RNet appearance representation | ER3 event recognition · CCV | mAP: 79.9 -> 87.1 | Helped |
| Use a larger number of reasoning-network hops (ablation) | ER3 event recognition · EVVE | test error | Helped |
| Add average spatial pooling inside the memory layer (ablation) | ER3 recounting · event recounting examples | recounting-map smoothness | Helped |
| Increase the active-map threshold to filter short shots (ablation) | ER3 event retrieval · EVVE | mAP | Mixed |
| Increase the counting-grid size beyond 24 (ablation) | ER3 event retrieval · EVVE | mAP | No clear effect |
Closest papers by meaning
- Unified Embedding and Metric Learning for Zero-Exemplar Event Detection (CVPR 2017 · similarity 0.85)
- Learning Latent Super-Events to Detect Multiple Activities in Videos (CVPR 2018 · similarity 0.85)
- Video Event Extraction via Tracking Visual States of Arguments (AAAI 2023 · similarity 0.84)
- VidEvent: A Large Dataset for Understanding Dynamic Evolution of Events in Videos (AAAI 2025 · similarity 0.84)
- MomentSeeker: A Task-Oriented Benchmark For Long-Video Moment Retrieval (NeurIPS 2025 · similarity 0.84)
- EventLens: Event-Structure Reinforcement Learning for Video Understanding (NeurIPS 2026 · similarity 0.84)
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.