Research map / Video / Video-language understanding
UniAV: Unified Audio-Visual Perception for Multi-Task Video Event Localization
Tiantian Geng, Teng Wang, Jinming Duan, Yanfu Zhang, Weili Guan, Feng Zheng, Ling Shao
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 |
|---|---|---|---|
| add task-specific experts and language-aware classifier to naive multi-task model | naive multi-task model · All three tasks | all-tasks average mAP: 47.7 -> 48.9 | Improved |
| train all tasks jointly with the unified model instead of single-task models | UniAV · DESED (SED) | average mAP: 57.7 -> 59.8 | Improved |
| fine-tune single-task models from the all-task model | UniAV AT · All three tasks | all-tasks average mAP: 48.9 -> 49.7 | Improved |
| naive multi-task training instead of single-task | single-task models · ActivityNet 1.3 (TAL) | average mAP: 35.3 -> 34.8 | Hurt |
| cross-task transfer of TAL model to AVEL (head fine-tuning only) | single-task TAL model · UnAV-100 (AVEL) | average mAP: 49.6 -> 29.8 | Hurt |
| jointly train SED with AVEL | UniAV · DESED (SED) | average mAP: 57.7 -> 59.4 | Improved |
| jointly train SED with AVEL | UniAV · UnAV-100 (AVEL) | average mAP: 49.6 -> 48.3 | Hurt |
| replace separate classifiers with language-aware classification head (ablation) | UniAV · DESED (SED) | average mAP: 54.8 -> 59.7 | Improved |
| add task-specific experts to later blocks rather than early blocks (ablation) | UniAV · UnAV-100 (AVEL) | average mAP: 50.5 -> 50.8 | Improved |
| add task-specific experts instead of widening dimension to 768 (ablation) | UniAV w/o experts · ActivityNet 1.3 (TAL) | average mAP: 34.9 -> 36.1 | Improved |
| add audio to visual input for TAL (ablation) | ST TAL model · ActivityNet 1.3 (TAL) | average mAP: 34.2 -> 35.3 | Improved |
| add visual to audio input for SED (ablation) | ST SED model · DESED (SED) | average mAP: 49.5 -> 57.7 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- TE-TAD — TE-TAD: Towards Full End-to-End Temporal Action Detection via Time-Aligned Coordinate Expression (CVPR 2024) · better on 3 of 3 shared results
- TriDet — TriDet: Temporal Action Detection With Relative Boundary Modeling (CVPR 2023) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Towards Open-Vocabulary Audio-Visual Event Localization (CVPR 2025 · similarity 0.92)
- Dense Audio-Visual Event Localization Under Cross-Modal Consistency and Multi-Temporal Granularity Collaboration (AAAI 2025 · similarity 0.91)
- Adapting to the Unknown: Training-Free Audio-Visual Event Perception with Dynamic Thresholds (CVPR 2025 · similarity 0.90)
- Prompting Segmentation with Sound Is Generalizable Audio-Visual Source Localizer (AAAI 2024 · similarity 0.89)
- Empowering LLMs with Pseudo-Untrimmed Videos for Audio-Visual Temporal Understanding (AAAI 2025 · similarity 0.89)
- Audio-Visual Semantic Graph Network for Audio-Visual Event Localization (CVPR 2025 · 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.
