Research map / Video / Action recognition
Knowledge is Power: Advancing Few-shot Action Recognition with Multimodal Semantics from MLLMs
Jiazheng Xing, Chao Xu, Hangjie Yuan, Mengmeng Wang, Jun Dan, Hangwei Qian, Yong Liu
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 |
|---|---|---|---|
| use hidden features directly instead of caption then encode (ablation) | FSAR-LLaVA (Unknown) · HMDB51 5-way 1-shot | accuracy: 69.8 -> 79.5 | Improved |
| use hidden features directly instead of caption then encode (ablation) | FSAR-LLaVA (Unknown) · Kinetics 5-way 1-shot | accuracy: 85.7 -> 88.1 | Improved |
| decouple visual and textual branches before pooling (ablation) | Video-LLaVA features · Kinetics 5-way 1-shot | accuracy: 79.3 -> 82.1 | Improved |
| add spatiotemporal self-attention in MFM (ablation) | FSAR-LLaVA (Unknown) · SSv2-Small 5-way 1-shot | accuracy: 48.5 -> 49.5 | Improved |
| add text-lead cross-attention in MFM (ablation) | FSAR-LLaVA (Known) · SSv2-Full 5-way 1-shot | accuracy: 65.2 -> 69.4 | Improved |
| add local prototype construction (ablation) | FSAR-LLaVA (Unknown) · Kinetics 5-way 1-shot | accuracy: 90.1 -> 94.5 | Improved |
| add global prototype construction to local (ablation) | FSAR-LLaVA (Known) · SSv2-Small 5-way 1-shot | accuracy: 60.6 -> 65.5 | Improved |
| use MPMM instead of Bi-MHM (ablation) | FSAR-LLaVA (Unknown) · SSv2-Small 5-way 1-shot | accuracy: 48.9 -> 51.0 | Improved |
| use MPMM instead of Hausdorff distance (ablation) | FSAR-LLaVA (Unknown) · Kinetics 5-way 1-shot | accuracy: 92.0 -> 94.8 | Improved |
| increase mapping dimension from 128 to 256 (ablation) | FSAR-LLaVA (Unknown) · Kinetics 5-way 1-shot | accuracy: 94.0 -> 94.8 | Improved |
| raise alpha from 0.1 to 0.9 (more local prototype weight) (ablation) | FSAR-LLaVA (Unknown) · SSv2-Small vs Kinetics 5-way 1-shot | accuracy: 51.0 -> 49.2 | Mixed |
| switch from Unknown to Known prompts | FSAR-LLaVA · SSv2-Small 5-way 1-shot | accuracy: 51.0 -> 65.5 | Improved |
Closest papers by meaning
- LLaVAction: evaluating and training multi-modal large language models for action understanding (ICLR 2026 · similarity 0.90)
- Weakly Supervised Temporal Action Localization via Dual-Prior Collaborative Learning Guided by Multimodal Large Language Models (CVPR 2025 · similarity 0.88)
- Frame Order Matters: A Temporal Sequence-Aware Model for Few-Shot Action Recognition (AAAI 2025 · similarity 0.88)
- A Multimodal, Multi-Task Adapting Framework for Video Action Recognition (AAAI 2024 · similarity 0.87)
- MMDiff: Multimodal Model Diffing for Feature Discovery and Control (NeurIPS 2026 · similarity 0.87)
- Context-Aware Multimodal Pretraining (CVPR 2025 · similarity 0.87)
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.
