Research map / Video / Action recognition
AVA: A Video Dataset of Spatio-Temporally Localized Atomic Visual Actions
Read the paperCheck 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 3D spatiotemporal convolutions rather than the 2D baseline (ablation) | Action localization model · JHMDB split1 | Frame-mAP @ IoU 0.5: 52.1% -> 67.9% | Helped |
| Use 3D spatiotemporal convolutions rather than the 2D baseline (ablation) | Action localization model · UCF101-24 split1 | Frame-mAP @ IoU 0.5: 60.1% -> 76.1% | Helped |
| Use 3D spatiotemporal convolutions rather than the 2D baseline (ablation) | Action localization model · AVA | Frame-mAP @ IoU 0.5: 13.7% -> 13.6% | No clear effect |
| Increase temporal input from 5 to 20 RGB + Flow frames (ablation) | 3D action localization model · AVA | Frame-mAP @ IoU 0.5: 13.6% -> 15.2% | Helped |
| Increase temporal input from 20 to 40 RGB + Flow frames (ablation) | 3D action localization model · AVA | Frame-mAP @ IoU 0.5: 15.2% -> 15.6% | Helped |
| Increase temporal input from 40 to 50 RGB + Flow frames (ablation) | 3D action localization model · AVA | Frame-mAP @ IoU 0.5: 15.6% -> 15.5% | No clear effect |
| Increase temporal input from 20 to 40 RGB + Flow frames (ablation) | 3D action localization model · UCF101-24 split1 | Frame-mAP @ IoU 0.5: 78.3% -> 76.0% | Hurt |
| Add RGB input to an optical-flow-only model (ablation) | 3D action localization model · AVA | Frame-mAP @ IoU 0.5: 9.9% -> 15.2% | Helped |
| Add optical-flow input to an RGB-only model (ablation) | 3D action localization model · AVA | Frame-mAP @ IoU 0.5: 14.5% -> 15.2% | Helped |
| Use class-agnostic actor localization rather than end-to-end action detection (ablation) | Our action localization model · AVA | Frame-mAP @ IoU 0.5: 15.6% -> 75.3% | Helped |
Closest papers by meaning
- UCF-STAR: A Large Scale Still Image Dataset for Understanding Human Actions (AAAI 2020 · similarity 0.90)
- VA-AR: Learning Velocity-Aware Action Representations with Mixture of Window Attention (AAAI 2025 · similarity 0.89)
- Storyboard-guided Alignment for Fine-grained Video Action Recognition (NeurIPS 2025 · similarity 0.88)
- Building a Multi-modal Spatiotemporal Expert for Zero-shot Action Recognition with CLIP (AAAI 2025 · similarity 0.88)
- Action-slot: Visual Action-centric Representations for Multi-label Atomic Activity Recognition in Traffic Scenes (CVPR 2024 · similarity 0.88)
- Unsupervised Action Segmentation via Fast Learning of Semantically Consistent Actoms (AAAI 2024 · similarity 0.88)
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.