Research map / Video / Video-language understanding
Dense-Localizing Audio-Visual Events in Untrimmed Videos: A Large-Scale Benchmark and Baseline
Tiantian Geng, Teng Wang, Jinming Duan, Runmin Cong, Feng Zheng
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 audio and visual input together rather than visual input alone (ablation) | Ours · UnAV-100 test | average mAP: 35.9 -> 47.8 | Improved |
| use the proposed audio-visual model instead of ActionFormer | UnAV-100 · UnAV-100 test | average mAP: 42.2 -> 47.8 | Improved |
| add temporal downsampling across cross-modal pyramid levels (ablation) | Cross-modal transformer · UnAV-100 test | average mAP: 35.3 -> 39.4 | Improved |
| increase cross-modal pyramid levels from two to six with temporal downsampling (ablation) | Cross-modal transformer · UnAV-100 test | average mAP: 39.4 -> 47.8 | Improved |
| increase cross-modal pyramid levels from six to seven (ablation) | Cross-modal transformer · UnAV-100 test | average mAP: 47.8 -> 46.8 | Hurt |
| add two uni-modal transformer blocks before cross-modal interaction (ablation) | Cross-modal transformer · UnAV-100 test | average mAP: 46.7 -> 47.8 | Improved |
| add temporal dependency modeling (ablation) | Dense-localization model · UnAV-100 test | average mAP: 45.2 -> 46.1 | Improved |
| add class-aware regression (ablation) | Dense-localization model · UnAV-100 test | average mAP: 45.2 -> 45.8 | Improved |
| use dependency modeling and class-aware regression together (ablation) | Dense-localization model · UnAV-100 test | average mAP: 45.2 -> 47.8 | Improved |
| use I3D RGB and optical-flow features instead of RGB-only I3D features (ablation) | Visual feature extractor · UnAV-100 test | average mAP: 46.7 -> 47.8 | Improved |
| add position embeddings (ablation) | Transformer encoder · UnAV-100 test | average mAP: 47.0 -> 47.8 | Improved |
| increase temporal feature stride from 8 to 16 (ablation) | Audio-visual feature extraction · UnAV-100 test | average mAP: 47.8 -> 46.7 | Hurt |
Closest papers by meaning
- Dense Audio-Visual Event Localization Under Cross-Modal Consistency and Multi-Temporal Granularity Collaboration (AAAI 2025 · similarity 0.95)
- Towards Open-Vocabulary Audio-Visual Event Localization (CVPR 2025 · similarity 0.91)
- Adapting to the Unknown: Training-Free Audio-Visual Event Perception with Dynamic Thresholds (CVPR 2025 · similarity 0.90)
- Empowering LLMs with Pseudo-Untrimmed Videos for Audio-Visual Temporal Understanding (AAAI 2025 · similarity 0.90)
- Audio-Visual Semantic Graph Network for Audio-Visual Event Localization (CVPR 2025 · similarity 0.89)
- Prompting Segmentation with Sound Is Generalizable Audio-Visual Source Localizer (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.
