Research map / Video / Video-language understanding
FPS-Bench: A Benchmark for High Frame-Rate Video Understanding
Rohan Choudhury, Jean-Sebastien Dandurand, Kai Qiu, Kshitij Madhav Bhat, Kartik Sharma, Liza Dahiya, Yizhou Zhao, Souraja Kundu, Chun-Hsien Lin, Kris M. Kitani, László A. Jeni
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 |
|---|---|---|---|
| slow videos down to ground-truth minFPS and trim to temporal certificate (ablation) | Gemini 2.5 Pro · FPS-Bench | accuracy: 28.9% -> 33.7% | Improved |
| slow videos down to minFPS plus temporal certificate (ablation) | Qwen-3-32B-instruct · FPS-Bench | accuracy: 30.7% -> 33.7% | Improved |
| slow videos down to minFPS plus temporal certificate (ablation) | GPT-4o · FPS-Bench | accuracy: 31.8% -> 32.1% | No clear effect |
| slow videos down to minFPS plus temporal certificate (ablation) | LLaVA-NeXT-Video · FPS-Bench | accuracy: 26.5% -> 26.0% | No clear effect |
| pass only the temporal certificate instead of the full 10-second clip (ablation) | VLMs · FPS-Bench | accuracy: GPT-4o 31.8% -> 32.2% | No clear effect |
| raise input FPS from 1 to 30 (ablation) | Qwen-3-32B-instruct · FPS-Bench | accuracy: 27.10% -> 35.20% | Improved |
| raise input FPS from 1 to 30 (ablation) | InternVL-3.5-14B · FPS-Bench | accuracy: 30.40% -> 26.70% | Hurt |
| raise input FPS from 1 to 30 (ablation) | Oryx · FPS-Bench | accuracy: 29.40% -> 30.50% | No clear effect |
| increase number of frames from 16 to 512 (ablation) | InternVL-3.5-8B · FPS-Bench | accuracy: 33.10% -> 26.50% | Hurt |
| increase number of frames from 16 to 512 (ablation) | LLaVA-OneVision · FPS-Bench | accuracy: 31.50% -> 27.20% | Hurt |
| increase number of frames from 16 to 128 (ablation) | OmniVinci · FPS-Bench | accuracy: 30.80% -> 32.60% | No clear effect |
| scale Qwen-3-VL from 8B to 32B (ablation) | Qwen-3-VL · FPS-Bench | accuracy: 28.0% -> 30.7% | Improved |
Closest papers by meaning
- Q-Bench-Video: Benchmark the Video Quality Understanding of LMMs (CVPR 2025 · similarity 0.89)
- CG-Bench: Clue-grounded Question Answering Benchmark for Long Video Understanding (ICLR 2025 · similarity 0.88)
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.88)
- RTV-Bench: Benchmarking MLLM Continuous Perception, Understanding and Reasoning through Real-Time Video (NeurIPS 2025 · similarity 0.88)
- E.T. Bench: Towards Open-Ended Event-Level Video-Language Understanding (NeurIPS 2024 · similarity 0.88)
- SVBench: A Benchmark with Temporal Multi-Turn Dialogues for Streaming Video Understanding (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.
