Research map / Video / Video-language understanding
Env-QA: A Video Question Answering Benchmark for Comprehensive Understanding of Dynamic Environments
Difei Gao, Ruiping Wang, Ziyi Bai, Xilin Chen
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 multi-step attention (ablation) | TSEA · Env-QA test split | overall accuracy (%): 45.46 -> 47.06 | Improved |
| add event-level feature (temporal segmentation) (ablation) | TSEA without multi-step attention · Env-QA test split | overall accuracy (%): 43.39 -> 45.46 | Improved |
| add focus attention (ablation) | TSEA without event feature · Env-QA test split | overall accuracy (%): 41.80 -> 43.39 | Improved |
| add object name feature (ablation) | TSEA without focus attention · Env-QA test split | overall accuracy (%): 40.60 -> 41.80 | Improved |
| evaluate on unseen environments instead of seen environments (ablation) | TSEA · Env-QA test split | accuracy (%): 46.85 -> 47.20 | No clear effect |
| use TSEA instead of STAGE | video QA model · Env-QA test split | overall accuracy (%): 42.53 -> 47.06 | Improved |
| add visual input to question (CNN-LSTM versus Q-ONLY) | video QA baseline · Env-QA test split | overall accuracy (%): 32.48 -> 38.05 | Improved |
| use TSEA instead of Q-ONLY on Event questions | video QA · Env-QA test split, Event questions | accuracy (%): 24.26 -> 39.84 | Improved |
| use TSEA instead of Q-ONLY on Number questions | video QA · Env-QA test split, Number questions | accuracy (%): 37.84 -> 39.35 | No clear effect |
| use TSEA instead of Q-ONLY on Order questions | video QA · Env-QA test split, Order questions | accuracy (%): 51.79 -> 55.53 | No clear effect |
| use TSEA instead of STAGE on State questions | video QA · Env-QA test split, State questions | accuracy (%): 49.93 -> 56.73 | Improved |
| use TSEA instead of STAGE on Action role | video QA · Env-QA test split, Action role | role accuracy (%): 45.69 -> 47.61 | No clear effect |
Closest papers by meaning
- Commonsense Video Question Answering through Video-Grounded Entailment Tree Reasoning (CVPR 2025 · similarity 0.88)
- Grounded Question-Answering in Long Egocentric Videos (CVPR 2024 · similarity 0.87)
- MoReVQA: Exploring Modular Reasoning Models for Video Question Answering (CVPR 2024 · similarity 0.87)
- Searching Videos as Trees: Self-Correcting Agents for Grounded Long Video QA (NeurIPS 2026 · similarity 0.87)
- Compositional 4D Dynamic Scenes Understanding with Physics Priors for Video Question Answering (ICLR 2025 · similarity 0.87)
- VideoEspresso: A Large-Scale Chain-of-Thought Dataset for Fine-Grained Video Reasoning via Core Frame Selection (CVPR 2025 · 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.
