Research map / Video / Video-language understanding
PPLLaVA: Varied Video Sequence Understanding With Prompt Guidance
Shangkun Sun, Ruyang Liu, Haoran Tang, Yixiao Ge, Haibo Lu, Jiankun Yang, Chen Li
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 prompt-guided pooling (1024 tokens) in place of average pooling (ablation) | LLaVA-Next · Video-MME (w/ subs) | accuracy: 43.4 -> 48.9 | Improved |
| add prompt-guided pooling instead of feeding all 4608 tokens (ablation) | LLaVA-Next · Video-MME (w/ subs) | accuracy: 47.4 -> 48.9 | Improved |
| add CLIP context extension (ablation) | PPLLaVA (LLaVA-Next) · Video-MME (w/ subs) overall | accuracy: 48.9 -> 50.0 | Improved |
| add CLIP context extension (ablation) | PPLLaVA (LLaVA-Next) · VCG-Bench average | score: 3.21 -> 3.32 | Improved |
| use weighted average pooling instead of separate spatiotemporal pooling (ablation) | PPLLaVA · Video-MME (w/ subs) | accuracy: 44.1 -> 53.6 | Improved |
| use weighted average pooling instead of max pooling (ablation) | PPLLaVA · Video-MME (w/ subs) | accuracy: 52.0 -> 53.6 | Improved |
| use weighted average pooling instead of Token Merging (ablation) | PPLLaVA · Video-MME (w/ subs) | accuracy: 51.9 -> 53.6 | Improved |
| use multiple pooling kernels instead of a single weighted-average kernel (ablation) | PPLLaVA · Video-MME (w/ subs) | accuracy: 53.6 -> 53.2 | No clear effect |
| manually select key frames instead of average sampling on redundant videos | PPLLaVA · Video-MME-redund | accuracy: 47.6 -> 50.5 | Improved |
| increase pooling kernel and stride in temporal dimension | PPLLaVA · Video-MME | accuracy | Hurt |
| increase spatial pooling kernel and stride | PPLLaVA · Video-MME | accuracy | No clear effect |
| apply PPLLaVA pooling to image LLM | LLaVA-Next-7B · MMMU (val) | accuracy: 35.8 -> 37.9 | Improved |
Closest papers by meaning
- ReWind: Understanding Long Videos with Instructed Learnable Memory (CVPR 2025 · similarity 0.90)
- Streaming Long Video Understanding with Large Language Models (NeurIPS 2024 · similarity 0.88)
- FastVID: Dynamic Density Pruning for Fast Video Large Language Models (NeurIPS 2025 · similarity 0.88)
- Unleashing Hour-Scale Video Training for Long Video-Language Understanding (NeurIPS 2025 · similarity 0.88)
- Video-XL: Extra-Long Vision Language Model for Hour-Scale Video Understanding (CVPR 2025 · similarity 0.88)
- TimeSuite: Improving MLLMs for Long Video Understanding via Grounded Tuning (ICLR 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.
