Research map / Video / Video-language understanding
LongVLM: Efficient Long Video Understanding via Large Language Models
Yuetian Weng, Mingfei Han, Haoyu He, Xiaojun Chang, Bohan Zhuang
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 3D average pooling with hierarchical token merging for local features (ablation) | LongVLM · Video-ChatGPT benchmark | Mean score: 2.64 -> 2.72 | Improved |
| replace 3D average pooling with token merging (ablation) | LongVLM (local only) · Egoschema | accuracy: 48.0 -> 52.4 | Improved |
| add global [CLS] semantic features to local features (ablation) | LongVLM · Video-ChatGPT benchmark | Mean score: 2.72 -> 2.89 | Improved |
| add global features on Egoschema (ablation) | LongVLM · Egoschema | accuracy: 52.4 -> 57.6 | Improved |
| prepend global feature before local features instead of appending (ablation) | LongVLM · Video-ChatGPT benchmark | Mean score: 2.82 -> 2.89 | Improved |
| increase tokens per segment M from 10 to 30 (ablation) | LongVLM · Video-ChatGPT benchmark | Mean score: 2.72 -> 2.89 | Improved |
| increase M from 30 to 40 (ablation) | LongVLM · ANET-QA | accuracy: 47.6 -> 46.0 | Hurt |
| increase M from 30 to 40 (ablation) | LongVLM · ANET-QA | GPU memory (G): 14.86 -> 14.96 | Hurt |
| use CLS tokens from last 5 layers rather than 1 (ablation) | LongVLM · Video-ChatGPT benchmark | Mean score: 2.83 -> 2.89 | Improved |
| increase E from 5 to 24 CLS layers (ablation) | LongVLM · Video-ChatGPT benchmark | Mean score: 2.89 -> 2.70 | Hurt |
| raise LLM temperature from 0.2 to 2.0 (ablation) | LongVLM · Video-ChatGPT benchmark | Consistency: 3.11 -> 1.01 | Hurt |
| lower temperature from 0.2 to 0.01 (ablation) | LongVLM · Video-ChatGPT benchmark | Temporal Understanding: 2.39 -> 2.22 | Hurt |
Later papers that compared with this method
1 later accepted paper reports LongVLM as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Streaming Long Video Understanding with Large Language Models (NeurIPS 2024 · better on 2 of 2 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BT-Adapter — BT-Adapter: Video Conversation is Feasible Without Video Instruction Tuning (CVPR 2024) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Streaming Long Video Understanding with Large Language Models (NeurIPS 2024 · similarity 0.93)
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.93)
- Leveraging Textual Memory and Key Frame Reasoning for Full Video Understanding Using Off-the-Shelf LLMs and VLMs (Student Abstract) (AAAI 2025 · similarity 0.92)
- MA-LMM: Memory-Augmented Large Multimodal Model for Long-Term Video Understanding (CVPR 2024 · similarity 0.92)
- Temporal Chain of Thought: Long-Video Understanding by Thinking in Frames (NeurIPS 2025 · similarity 0.92)
- ReWind: Understanding Long Videos with Instructed Learnable Memory (CVPR 2025 · similarity 0.91)
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.
