Research map / Video / Video-language understanding
Open-vocabulary Video Question Answering: A New Benchmark for Evaluating the Generalizability of Video Question Answering Models
Dohwan Ko, Ji Soo Lee, Miso Choi, Jaewon Chu, Jihwan Park, Hyunwoo J. Kim
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 |
|---|---|---|---|
| replace classification head with answer encoder similarity (OVQA baseline) | FrozenBiLM · MSVD-QA | mAcc: 17.2 -> 21.7 | Improved |
| replace classification head with answer encoder similarity (OVQA baseline) | VIOLET · MSVD-QA | mAcc: 2.7 -> 10.7 | Improved |
| replace classification head with answer encoder similarity (OVQA baseline) | baselines · all four datasets | base-answer accuracy: FrozenBiLM 72.7 -> 72.2 (MSVD B) | Mixed |
| add GNN-based soft verbalizer (ablation) | FrozenBiLM+ · MSVD-QA | unseen accuracy: 13.7 -> 16.1 | Improved |
| add GNN-based soft verbalizer (ablation) | FrozenBiLM+ · ActivityNet-QA | mAcc: 10.4 -> 12.4 | Improved |
| add GNN-based soft verbalizer (ablation) | All-in-one+ · ActivityNet-QA | mAcc: 6.8 -> 8.7 | Improved |
| add GNN-based soft verbalizer (ablation) | JustAsk+ · ActivityNet-QA | common accuracy: 37.1 -> 35.6 | Mixed |
| use a hard verbalizer (mean-pooling similar words) instead of no verbalizer (ablation) | FrozenBiLM+ · ActivityNet-QA | mAcc: 10.4 -> 7.9 | Hurt |
| use GNN-based soft verbalizer over learnable MLP verbalizer (ablation) | FrozenBiLM+ · ActivityNet-QA | mAcc: 10.8 -> 12.4 | Improved |
| use DeBERTa instead of CLIP as answer encoder (ablation) | All-in-one+ · MSVD-QA | mAcc: 5.3 -> 9.4 | Improved |
| vary convex combination coefficient epsilon from 1.0 to 0.7 (ablation) | FrozenBiLM+ GNN verbalizer · ActivityNet-QA | mAcc: 10.4 -> 11.9 | Improved |
| set epsilon to 0.0 (use only GNN outputs) (ablation) | FrozenBiLM+ GNN verbalizer · ActivityNet-QA | total accuracy: 43.2 -> 41.6 | Hurt |
Closest papers by meaning
- Invariant Grounding for Video Question Answering (CVPR 2022 · similarity 0.92)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.92)
- Evidence-Backed Video Question Answering (ECCV 2026 · similarity 0.92)
- Commonsense Video Question Answering through Video-Grounded Entailment Tree Reasoning (CVPR 2025 · similarity 0.91)
- MoReVQA: Exploring Modular Reasoning Models for Video Question Answering (CVPR 2024 · similarity 0.90)
- CG-Bench: Clue-grounded Question Answering Benchmark for Long Video Understanding (ICLR 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.
