Research map / Video / Video-language understanding
VideoQA in the Era of LLMs: An Empirical Study
Junbin Xiao, Nanxin Huang, Hangyu Qin, Dongyang Li, Yicong Li, Fengbin Zhu, Zhulin Tao, Jianxing Yu, Liang Lin, Tat‐Seng Chua, Angela Yao
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 TD-style data (about 4.5K) to NExT-QA training | FrozenGQA · NExT-QA temporal description (TD) | accuracy: 32.3 -> 92.5 | Improved |
| add TD-style data to NExT-QA training | FrozenGQA · NExT-QA temporal exchange (TE) | accuracy: 46.1 -> 52.0 | Improved |
| add TD-style data to NExT-QA training | FrozenGQA · NExT-QA temporal original (TO) | accuracy: 67.7 -> 68.3 | Improved |
| prompt GPT-4o to also output indexes of frames it uses | GPT-4o · NExT-GQA NormalVQA | accuracy: 49.0 -> 64.7 | Improved |
| remove video input (BlindQA) for finetuned vs zero-shot models (ablation) | CoVGT vs GPT-4 · NExT-GQA BlindQA | accuracy: 49.3 vs 45.3 | Mixed |
| rewrite answer options to remove question-answer and video-answer shortcuts | Video-LLMs · NExT-QA validation subset | accuracy: 10% ~ 20% decline | Hurt |
| edit options to remove video-answer shortcut vs question-answer shortcut | GPT-4o · NExT-QA | accuracy drop: 2.6% (QA) vs 9.9% (VA) | Mixed |
| add spoken phrases before the question | GPT-4o · NExT-QA | accuracy: +3% | Improved |
| shuffle video frames | Video-LLMs · NExT-QA and ActivityNet-QA | accuracy | No clear effect |
| shuffle question words | Video-LLMs · MCQA and OEQA | accuracy retained: 86%~97% (MCQA), 70%~80% (OEQA) | Hurt |
| scale LLaMA-VID LLM from 7B to 13B (ablation) | LLaMA-VID · OEQA with spoken prefixes | accuracy | Improved |
| train on temporal or causal questions instead of descriptive questions for testing on descriptive | SeViLA and LLaMA-VQA · NExT-QA descriptive questions | accuracy | No clear effect |
Closest papers by meaning
- Can Video LLMs Refuse to Answer? Alignment for Answerability in Video Large Language Models (ICLR 2025 · similarity 0.93)
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.91)
- On the Consistency of Video Large Language Models in Temporal Comprehension (CVPR 2025 · similarity 0.90)
- Modularized Self-Reflected Video Reasoner for Multimodal LLM with Application to Video Question Answering (ICML 2025 · similarity 0.90)
- MoReVQA: Exploring Modular Reasoning Models for Video Question Answering (CVPR 2024 · similarity 0.90)
- Temporal Chain of Thought: Long-Video Understanding by Thinking in Frames (NeurIPS 2025 · similarity 0.90)
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.
