Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Learning to Compose Topic-Aware Mixture of Experts for Zero-Shot Video Captioning
Xin Wang, Jiawei Wu, Da Zhang, Yu Su, William Yang Wang
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 |
|---|---|---|---|
| Compose topic-conditioned primitive experts instead of using the Base captioner | ActivityNet-Captions zero-shot video captioning · ActivityNet-Captions unseen test set | CIDEr: 22.51 -> 28.23 | Improved |
| Add topic-conditioned expert composition to the Base model | ActivityNet-Captions captioning · ActivityNet-Captions seen test set | CIDEr: 31.48 -> 35.53 | Improved |
| Add pretrained fasttext embeddings instead of task-specific embeddings | TAMoE · ActivityNet-Captions unseen test set | CIDEr: 24.39 -> 28.23 | Improved |
| Add the topic embedding to the Base model | ActivityNet-Captions zero-shot video captioning · ActivityNet-Captions unseen test set | CIDEr: 22.51 -> 24.06 | Improved |
| Add the average activity-label embedding to I3D video features (ablation) | TAMoE · ActivityNet-Captions unseen test set | CIDEr: 22.51 -> 25.96 | Improved |
| Add the external-corpus TFIDF embedding to I3D video features (ablation) | TAMoE · ActivityNet-Captions unseen test set | CIDEr: 22.51 -> 26.61 | Improved |
| Combine average label and TFIDF embeddings with I3D video features (ablation) | TAMoE · ActivityNet-Captions unseen test set | CIDEr: 22.51 -> 28.23 | Improved |
| Remove I3D video features (ablation) | TAMoE · ActivityNet-Captions unseen test set | CIDEr: 28.23 -> 15.77 | Hurt |
| Use a single expert instead of a multi-expert mixture (ablation) | TAMoE · ActivityNet-Captions validation set | CIDEr | Hurt |
| Increase the expert count to 256 with expert dimension 128 rather than using 8 experts of dimension 256 (ablation) | TAMoE · ActivityNet-Captions validation set | CIDEr | Hurt |
| Use topic-aware mixture-of-experts captioning rather than Base | MSR-VTT · MSR-VTT | CIDEr: 47.2 -> 48.9 | Improved |
| Use TAMoE rather than Base for individual novel activities | ActivityNet-Captions · Novel activities | CIDEr | Mixed |
Closest papers by meaning
- Set Prediction Guided by Semantic Concepts for Diverse Video Captioning (AAAI 2024 · similarity 0.86)
- Exploiting Auxiliary Caption for Video Grounding (AAAI 2024 · similarity 0.86)
- Zero-Shot Image Captioning with Multi-type Entity Representations (AAAI 2025 · similarity 0.86)
- Learnability Matters: Active Learning for Video Captioning (NeurIPS 2024 · similarity 0.86)
- Comprehensive Visual Grounding for Video Description (AAAI 2024 · similarity 0.85)
- Shot2Story: A New Benchmark for Comprehensive Understanding of Multi-shot Videos (ICLR 2025 · 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.
