Research map / Video / Video-language understanding
KnowIT VQA: Answering Knowledge-Based Questions about Videos
Noa Garcia, Mayu Otani, Chenhui Chu, Yuta Nakashima
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 retrieved show-specific knowledge to video, subtitles, questions, and answers | ROCK · KnowIT VQA test (All) | accuracy: 0.587 -> 0.652 | Improved |
| Replace retrieved knowledge with annotated ground-truth knowledge (ablation) | ROCK · KnowIT VQA test (All) | accuracy: 0.652 -> 0.731 | Improved |
| Use caption features instead of image features (ablation) | ROCK · KnowIT VQA test (All) | accuracy: 0.652 -> 0.646 | Hurt |
| Use concepts, facial features, or image features in place of one another (ablation) | ROCK · KnowIT VQA test (All) | accuracy: 0.652 | No clear effect |
| Include subtitles with questions and answers (ablation) | ROCK_QA · KnowIT VQA test (All) | accuracy: 0.530 -> 0.587 | Improved |
| Add visual representations to subtitles, questions, and answers (ablation) | ROCK_SQA · KnowIT VQA test (All) | accuracy: 0.587 -> 0.587 | No clear effect |
| Add subtitles using LSTM Embeddings (ablation) | LSTM Emb. · KnowIT VQA test (All) | accuracy: 0.489 -> 0.467 | Hurt |
| Use candidate answers together with the question for knowledge retrieval (ablation) | Knowledge retrieval module · KnowIT VQA knowledge retrieval test | R@1: 0.070 -> 0.083 | Improved |
| Use all candidate answers together, ordered by prior score, rather than one at a time (ablation) | Knowledge retrieval module · KnowIT VQA knowledge retrieval test | R@1: 0.083 -> 0.114 | Improved |
| Provide annotated knowledge to rookies answering questions | Rookies answering with subtitles and video · KnowIT VQA test | accuracy: 0.683 -> 0.823 | Improved |
Closest papers by meaning
- Knowledge-Based Video Question Answering with Unsupervised Scene Descriptions (ECCV 2020 · similarity 0.89)
- FriendsQA: A New Large-Scale Deep Video Understanding Dataset with Fine-grained Topic Categorization for Story Videos (AAAI 2025 · similarity 0.89)
- Commonsense Video Question Answering through Video-Grounded Entailment Tree Reasoning (CVPR 2025 · similarity 0.89)
- VideoEspresso: A Large-Scale Chain-of-Thought Dataset for Fine-Grained Video Reasoning via Core Frame Selection (CVPR 2025 · similarity 0.88)
- MoReVQA: Exploring Modular Reasoning Models for Video Question Answering (CVPR 2024 · similarity 0.88)
- Language-aware Visual Semantic Distillation for Video Question Answering (CVPR 2024 · 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.
