Research map / Vision-language and multimodal / Visual question answering
Segment-Then-Rank: Non-Factoid Question Answering on Instructional Videos
Kyungjae Lee, Nan Duan, Lei Ji, Jason Li, Seung-won Hwang
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 Ranker on top of Segmenter (two-stage) (ablation) | Segmenter · Instructional video non-factoid QA test set | F1: 65.71 -> 70.31 | Improved |
| Add object features to the encoder (ablation) | Segmenter · Instructional video non-factoid QA test set | recall: 64.45 -> 67.21 | Improved |
| Add object features to the encoder (ablation) | Segmenter + Ranker · Instructional video non-factoid QA test set | recall: 71.33 -> 74.75 | Improved |
| Add object features to the encoder (ablation) | Segmenter + Ranker · Instructional video non-factoid QA test set | F1: 68.53 -> 70.31 | Improved |
| Add self-attention and LSTM hierarchical modeling (ablation) | Segmenter · Instructional video non-factoid QA test set | F1: 51.31 -> 65.71 | Improved |
| Add LSTM on top of self-attention-free Segmenter (ablation) | Segmenter · Instructional video non-factoid QA test set | F1: 51.31 -> 64.08 | Improved |
| Add self-attention to Segmenter (ablation) | Segmenter · Instructional video non-factoid QA test set | F1: 64.08 -> 65.71 | Improved |
| Use interaction-based features instead of representation-based (ablation) | Segmenter + Ranker · Instructional video non-factoid QA test set | F1: 67.57 -> 68.57 | Improved |
| Concatenate representation and interaction features (ablation) | Segmenter + Ranker · Instructional video non-factoid QA test set | F1: 68.57 -> 68.74 | Improved |
| Use length-adaptive gate instead of concatenation (ablation) | Segmenter + Ranker · Instructional video non-factoid QA test set | F1: 68.74 -> 70.31 | Improved |
| Use Segmenter+Ranker instead of Ranker over fixed windows | Ranker · Instructional video non-factoid QA test set | F1: 53.97 -> 70.31 | Improved |
| Use BERT encoder with Segmenter instead of QANet encoder | Base-III · Instructional video non-factoid QA test set | F1: 61.57 -> 65.71 | Improved |
Closest papers by meaning
- Evidence-Backed Video Question Answering (ECCV 2026 · similarity 0.88)
- Leveraging Video Descriptions to Learn Video Question Answering (AAAI 2017 · similarity 0.88)
- CACR: Reinforcing Temporal Answer Grounding in Instructional Video via Candidate-Aware Causal Reasoning (ICML 2026 · similarity 0.87)
- Ranking Distillation for Open-Ended Video Question Answering with Insufficient Labels (CVPR 2024 · similarity 0.86)
- YTCommentQA: Video Question Answerability in Instructional Videos (AAAI 2024 · similarity 0.86)
- Question-Aware Gaussian Experts for Audio-Visual Question Answering (CVPR 2025 · similarity 0.86)
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.
