Research map / Video / Video-language understanding
Knowledge-Based Video Question Answering with Unsupervised Scene Descriptions
Noa Garcia, Yuta Nakashima
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 |
|---|---|---|---|
| use plot summaries as external knowledge instead of human annotations | ROLL · KnowIT VQA test (Knowledge) | accuracy: 0.567 -> 0.713 | Improved |
| use unsupervised scene-graph descriptions instead of captions for vision | ROLL-human · KnowIT VQA test (Visual) | accuracy: 0.647 -> 0.708 | Improved |
| full ROLL versus best previous ROCK | ROLL · KnowIT VQA test (All) | accuracy: 0.652 -> 0.715 | Improved |
| add recall branch to read branch (ablation) | Read · KnowIT VQA test (All) | accuracy: 0.584 -> 0.711 | Improved |
| add observe branch to read branch (ablation) | Read · KnowIT VQA test (All) | accuracy: 0.584 -> 0.590 | Improved |
| add observe branch to recall branch (ablation) | Recall · KnowIT VQA test (All) | accuracy: 0.685 -> 0.691 | Improved |
| add observe branch to read and recall branches (ablation) | Read-Recall · KnowIT VQA test (All) | accuracy: 0.711 -> 0.715 | Improved |
| add read branch to observe and recall branches (ablation) | Observe-Recall · KnowIT VQA test (All) | accuracy: 0.691 -> 0.715 | Improved |
| add modality weighting to linear fusion (ablation) | FC fusion · KnowIT VQA test (All) | accuracy: 0.669 -> 0.715 | Improved |
| fuse branches by average instead of modality weighting (ablation) | ROLL · KnowIT VQA test (All) | accuracy: 0.672 -> 0.715 | Hurt |
| add modality weighting weight beta_omega below 0.6 (ablation) | ROLL · KnowIT VQA test (All) | accuracy | Hurt |
| apply spatio-temporal filter to character recognition (ablation) | Character recognition · Character recognition (100 frames, F1) | F1: 87.96 -> 74.86 | Hurt |
Closest papers by meaning
- On the Hidden Treasure of Dialog in Video Question Answering (ICCV 2021 · similarity 0.90)
- MovieQA: Understanding Stories in Movies Through Question-Answering (CVPR 2016 · similarity 0.90)
- KnowIT VQA: Answering Knowledge-Based Questions about Videos (AAAI 2020 · similarity 0.89)
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.89)
- Temporal Chain of Thought: Long-Video Understanding by Thinking in Frames (NeurIPS 2025 · similarity 0.88)
- FriendsQA: A New Large-Scale Deep Video Understanding Dataset with Fine-grained Topic Categorization for Story Videos (AAAI 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.
