Research map / Video / Video-language understanding
HumanVBench: Probing Human-Centric Video Understanding in MLLMs with Automatically Synthesized Benchmarks
Ting Zhou, Daoyuan Chen, Qirui Jiao, Bolin Ding, Yaliang Li, Ying Shen
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 |
|---|---|---|---|
| restrict emotion recognition to speaking targets (ablation) | Video MLLMs · HumanVBench Emotion Recognition | accuracy (%): 36.9 -> 33.5 | Hurt |
| add textual timestamps to models without native support (ablation) | Video MLLMs · HumanVBench Appearance Time Detection | accuracy (%): average +6 (26.8) | Improved |
| add textual timestamps to models without native support (ablation) | Video MLLMs · HumanVBench Action at Specified Time | accuracy (%): average +4 (34.3) | Improved |
| add textual timestamps to models without native support (ablation) | Video MLLMs · HumanVBench Time of Specific Action | accuracy (%): average +2 (30.1); per-model from -4 to +19 | Mixed |
| use ensemble of MLLMs to create answers and distractors | Distractor-Included QA Synthesis Pipeline · HumanVBench Emotion Recognition | no-edit rate (%): 81% no-edit vs best model accuracy 49.4% | Improved |
| add audio input to MLLMs (audio-visual vs visual-only) | open-source MLLMs · HumanVBench Speech-Visual Alignment | accuracy (%): VideoLLaMA2.1 32.9 vs Qwen-VL3 48.3 | No clear effect |
| use proprietary Gemini-2.5-Pro instead of best open-source model | MLLM · HumanVBench (16 tasks) | overall accuracy (%): 53.4 -> 73.4 | Improved |
Closest papers by meaning
- MVBench: A Comprehensive Multi-modal Video Understanding Benchmark (CVPR 2024 · similarity 0.90)
- VideoAutoArena: An Automated Arena for Evaluating Large Multimodal Models in Video Analysis through User Simulation (CVPR 2025 · similarity 0.90)
- MMBench-Video: A Long-Form Multi-Shot Benchmark for Holistic Video Understanding (NeurIPS 2024 · similarity 0.89)
- Human-MME: A Holistic Evaluation Benchmark for Human-Centric Multimodal Large Language Models (ICLR 2026 · similarity 0.89)
- MMOU: A Massive Multi-Task Omni Understanding and Reasoning Benchmark for Long and Complex Real-World Videos (NeurIPS 2026 · similarity 0.89)
- FAVOR-Bench: A Comprehensive Benchmark for Fine-Grained Video Motion Understanding (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.
