Research map / Video / Video-language understanding
RIVER: Real-time Video Interaction Benchmark
Yansong Shi, Qingsong Zhao, Tianxiang Jiang, Xiangyu Zeng, Yi Wang, Limin Wang
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 RIVER pro-response fine-tuning data to VideoLLM-Online (2fps) | VideoLLM-Online · RIVER Pro-Response Instant | MC accuracy: 23.88 -> 33.28 | Improved |
| raise frame rate from 2fps to 4fps for VideoLLM-Online+RIVER | VideoLLM-Online+RIVER · RIVER Pro-Response Instant | MC accuracy: 33.28 -> 35.16 | Improved |
| add long-short term memory and sliding window to VideoChat-Flash | VideoChat-Flash · RIVER Retro-Memory | MC accuracy: 41.51 -> 45.75 | Improved |
| add long-short term memory and sliding window to InternVL2.5 | InternVL2.5 · RIVER Live-Perception | MC accuracy: 43.65 -> 58.84 | Improved |
| add long-short term memory and sliding window to InternVL2.5 | InternVL2.5 · RIVER Retro-Memory | MC accuracy: 42.68 -> 39.72 | Hurt |
| add long-short term memory and sliding window to Llava-Video | Llava-Video · RIVER Retro-Memory | MC accuracy: 46.00 -> 42.71 | Hurt |
| add long-short term memory and sliding window to Llava-Video | Llava-Video · RIVER Live-Perception | MC accuracy: 41.00 -> 51.38 | Improved |
| add long-short term memory and sliding window to VideoChat2 | VideoChat2 · RIVER Live-Perception | MC accuracy: 29.28 -> 41.16 | Improved |
| add memory module to MLLM for long recall questions (ablation) | MLLM (Figure 5) · RIVER Retro-Memory | decay slope: 12% lower decay slope | Improved |
| increase recall interval from short to very long | GPT-4o · RIVER Retro-Memory | MC accuracy: 63.26 -> 52.21 | Hurt |
| increase recall interval from short to very long | Llava-Video (16 frames) · RIVER Retro-Memory | MC accuracy: 49.17 -> 42.54 | Hurt |
| add online pipeline to VideoChat-Flash on background cues | VideoChat-Flash · RIVER cue types | MC accuracy (BC): 44.92 -> 54.10 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- VideoChat-Flash — VideoChat-Flash: Hierarchical Compression for Long-Context Video Modeling (ICLR 2026) · better on 3 of 3 shared results
- Flash-VStream — Flash-VStream: Efficient Real-Time Understanding for Long Video Streams (ICCV 2025) · better on 3 of 3 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
- RTV-Bench: Benchmarking MLLM Continuous Perception, Understanding and Reasoning through Real-Time Video (NeurIPS 2025 · similarity 0.88)
- Dispider: Enabling Video LLMs with Active Real-Time Interaction via Disentangled Perception, Decision, and Reaction (CVPR 2025 · similarity 0.88)
- Streaming Video Understanding and Multi-round Interaction with Memory-enhanced Knowledge (ICLR 2025 · similarity 0.88)
- StreamGaze: Gaze-Guided Temporal Reasoning and Proactive Understanding in Streaming Videos (NeurIPS 2026 · similarity 0.87)
- E.T. Bench: Towards Open-Ended Event-Level Video-Language Understanding (NeurIPS 2024 · similarity 0.87)
- MMBench-Video: A Long-Form Multi-Shot Benchmark for Holistic Video Understanding (NeurIPS 2024 · 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.
