Research map / Video / Video-language understanding
LongVideo-R1: Smart Navigation for Low-cost Long Video Understanding
Jihao Qiu, Lingxi Xie, Xinyue Huo, Qi Tian, Qixiang Ye
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 RL with GRPO after SFT (full data) (ablation) | LongVideo-R1 Qwen3-8B · LVBench | accuracy: 41.6 -> 50.0 | Improved |
| increase SFT data from 10K to full 33K samples (after RL) (ablation) | LongVideo-R1 · LVBench | accuracy: 47.4 -> 50.0 | Improved |
| increase SFT data from 10K to full (SFT only) (ablation) | LongVideo-R1-SFT · LVBench | accuracy: 39.1 -> 41.6 | Improved |
| add location reward r_loc in RL (ablation) | LongVideo-R1 · LVBench | accuracy: 45.8 -> 50.0 | Improved |
| add location reward r_loc in RL (ablation) | LongVideo-R1 · Video-MME long | accuracy: 61.4 -> 64.4 | Improved |
| add location reward r_loc in RL (ablation) | LongVideo-R1 · LVBench temporal grounding | accuracy: 53.2 -> 56.4 | Improved |
| scale captioning MLLM from 3B to 72B (ablation) | LongVideo-R1 · LVBench | accuracy: 44.5 -> 50.0 | Improved |
| scale captioning MLLM from 32B to 72B (ablation) | LongVideo-R1 · LVBench | accuracy: 49.8 -> 50.0 | No clear effect |
| increase maximum tool rounds from 10 to 30 (ablation) | LongVideo-R1 · LVBench | accuracy: 43.0 -> 50.0 | Improved |
| increase maximum tool rounds from 20 to 30 (ablation) | LongVideo-R1 · LVBench | accuracy: 47.3 -> 50.0 | Improved |
| use Qwen3-VL-32B captions and renewed SFT data | LongVideo-R1 · LVBench | accuracy: 50.0 -> 60.7 | Improved |
| use Qwen3-VL-32B captions and renewed SFT data | LongVideo-R1 · MLVU | accuracy: 68.1 -> 71.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 5 on most shared results.
- VideoChat-Flash-7B — VideoChat-Flash: Hierarchical Compression for Long-Context Video Modeling (ICLR 2026) · better on 1 of 2 shared results
- VideoTree — VideoTree: Adaptive Tree-based Video Representation for LLM Reasoning on Long Videos (CVPR 2025) · better on 2 of 2 shared results
- VideoMind-7B — VideoMind: A Chain-of-LoRA Agent for Temporal-Grounded Video Reasoning (ICLR 2026) · better on 1 of 1 shared result
- Ego-R1 — Ego-R1: Agentic Chain-of-Tool-Thought for Ultra-Long Egocentric Video Reasoning (TPAMI 2026) · better on 1 of 1 shared result
- VideoAgent — VideoAgent: A Memory-augmented Multimodal Agent for Video Understanding (ECCV 2024) · better on 1 of 1 shared result
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
- VideoZoomer: Reinforcement-Learned Temporal Focusing for Long Video Reasoning (ICLR 2026 · similarity 0.91)
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.89)
- Re-thinking Temporal Search for Long-Form Video Understanding (CVPR 2025 · similarity 0.89)
- Deep Video Discovery: Agentic Search with Tool Use for Long-form Video Understanding (NeurIPS 2025 · similarity 0.89)
- Understanding Long Videos with Multimodal Language Models (ICLR 2025 · similarity 0.89)
- Video-R1: Reinforcing Video Reasoning in MLLMs (NeurIPS 2025 · similarity 0.89)
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.
