Research map / Video / Video-language understanding
Leveraging Weak Semantic Relevance for Complex Video Event Classification
Chao Li, Jiewei Cao, Zi Huang, Lei Zhu, Heng Tao Shen
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 weak semantic relevance guidance to the attention model (lambda_a > 0) (ablation) | BiLSTM attention model · ActivityNet | mAP: 55.1% -> 61.6% | Improved |
| add weak semantic relevance guidance to the attention model (lambda_a > 0) (ablation) | BiLSTM attention model · FCVID | mAP: 73.2% -> 77.8% | Improved |
| add weak semantic relevance guidance to the attention model (lambda_a > 0) (ablation) | BiLSTM attention model · MEDTest14 | mAP: 29.1% -> 36.3% | Improved |
| use relevance as weak guidance in a learned attention model instead of directly weighting shot features (SVM-WA) | video event classifier · ActivityNet | mAP: 50.8% -> 61.6% | Improved |
| use relevance as weak guidance in a learned attention model instead of directly weighting shot features (SVM-WA) | video event classifier · FCVID | mAP: 69.9% -> 77.8% | Improved |
| use relevance as weak guidance in a learned attention model instead of directly weighting shot features (SVM-WA) | video event classifier · MEDTest14 | mAP: 28.1% -> 36.3% | Improved |
| use attention guidance plus bidirectional LSTM instead of NISVM isotonic ordering | video event classifier · MEDTest14 | mAP: 34.4% -> 36.3% | Improved |
| increase lambda_a above 0.4 (ablation) | full model · ActivityNet | mAP | Hurt |
| set lambda_a to 0.3 on FCVID (best value) (ablation) | full model · FCVID | mAP | Improved |
| add guided attention with LSTM on static-scene events (events 11 to 14) | full model · MEDTest14 events 11-14 | AP | Hurt |
Closest papers by meaning
- They Are Not Equally Reliable: Semantic Event Search Using Differentiated Concept Classifiers (CVPR 2016 · similarity 0.89)
- Video2vec Embeddings Recognize Events When Examples Are Scarce (TPAMI 2016 · similarity 0.88)
- Visual Semantic Role Labeling for Video Understanding (CVPR 2021 · similarity 0.88)
- Beyond Grounding: Extracting Fine-Grained Event Hierarchies across Modalities (AAAI 2024 · similarity 0.85)
- SEAL: Semantic Attention Learning for Long Video Representation (CVPR 2025 · similarity 0.85)
- Weakly-Supervised Audio-Visual Video Parsing with Prototype-based Pseudo-Labeling (CVPR 2024 · similarity 0.85)
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.
