Research map / Video / Video-language understanding
PVChat: Personalized Video Chat with One-Shot Learning
Yufei Shi, Weilong Yan, Gang Xu, Yumeng Li, Yucheng Chen, Zhenxi Li, Fei Yu, Ming Li, Si Yong Yeo
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 |
|---|---|---|---|
| replace Q-former baseline with ReMoH attention (ablation) | PVChat · personalized video QA | accuracy: 0.733 -> 0.901 | Improved |
| replace MoH with ReMoH (ablation) | PVChat · personalized video QA | accuracy: 0.813 -> 0.901 | Improved |
| add MoH to Q-former baseline (ablation) | PVChat · personalized video QA | accuracy: 0.733 -> 0.813 | Improved |
| add HAE loss alongside SPR (ablation) | PVChat · personalized video QA | accuracy: 0.746 -> 0.901 | Improved |
| add HAE loss alongside SPR (ablation) | PVChat · head activation rate | activation rate: 0.217 -> 0.552 | Improved |
| add SPR and HAE losses (ablation) | PVChat · training stability | loss: nan -> 0.028 | Improved |
| add negative samples to one-positive training data (ablation) | PVChat · personalized video QA | accuracy: 0.464 -> 0.584 | Improved |
| add ConsisID-generated positive videos (ablation) | PVChat · personalized video QA | accuracy: 0.584 -> 0.781 | Improved |
| add LivePortrait-generated positive videos (ablation) | PVChat · personalized video QA | accuracy: 0.781 -> 0.901 | Improved |
| increase tokens per character from 0 to 16 (ablation) | PVChat · personalized video QA | accuracy: 0.801 -> 0.922 | Improved |
| increase tokens per character from 16 to 20 (ablation) | PVChat · personalized video QA | accuracy: 0.922 -> 0.882 | Hurt |
| one-shot personalization versus VideoLLaMA2 with description | PVChat · personalized video QA | accuracy: 0.470 -> 0.901 | Improved |
Closest papers by meaning
- Evidence-Backed Video Question Answering (ECCV 2026 · similarity 0.89)
- MovieChat+: Question-Aware Sparse Memory for Long Video Question Answering (TPAMI 2025 · similarity 0.87)
- VideoChat-A1: Thinking with Long Videos by Chain-of-Shot Reasoning (AAAI 2026 · similarity 0.87)
- Online Video Understanding: OVBench and VideoChat-Online (CVPR 2025 · similarity 0.87)
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.87)
- Koala: Key Frame-Conditioned Long Video-LLM (CVPR 2024 · similarity 0.86)
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.
