Research map / Vision-language and multimodal / Visual question answering
Video as Conditional Graph Hierarchy for Multi-Granular Question Answering
Junbin Xiao, Angela Yao, Zhiyuan Liu, Yicong Li, Wei Ji, Tat-Seng Chua
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 the bottom object-level graph QGA G_O (ablation) | HQGA · NExT-QA Val | accuracy: 50.50 -> 51.42; drops by 0.9% when removed | Improved |
| add the bottom object-level graph QGA G_O (ablation) | HQGA · MSRVTT-QA Val | accuracy: 37.26 -> 38.23; drops by 0.9% when removed | Improved |
| add the middle-level graph QGA G_F for short-term dynamics (ablation) | HQGA · NExT-QA Val | accuracy: 50.00 -> 51.42; drops over 1.2% when removed | Improved |
| add the middle-level graph QGA G_F for short-term dynamics (ablation) | HQGA · MSRVTT-QA Val | accuracy: 37.05 -> 38.23; drops over 1.2% when removed | Improved |
| use a hierarchical graph instead of a monolithic clip-level graph (ablation) | HQGA · NExT-QA Val | accuracy: 49.96 -> 51.42; drops 1.5% | Improved |
| use a hierarchical graph instead of a monolithic clip-level graph (ablation) | HQGA · MSRVTT-QA Val | accuracy: 35.66 -> 38.23; drops 2.6% | Improved |
| use graph attention at the top level rather than sum pooling (ablation) | HQGA · NExT-QA Val | accuracy: 50.74 -> 51.42; drops 0.68% | Improved |
| use graph attention at the top level rather than sum pooling (ablation) | HQGA · MSRVTT-QA Val | accuracy: 37.69 -> 38.23; drops 0.54% | Improved |
| inject query conditioning at all levels rather than removing all query conditions (ablation) | HQGA · NExT-QA Val | accuracy: 50.62 -> 51.42 | Improved |
| inject query conditioning at all levels rather than removing all query conditions (ablation) | HQGA · MSRVTT-QA Val | accuracy: 37.03 -> 38.23 | Improved |
| use token-wise query conditions instead of a global query condition (ablation) | HQGA · NExT-QA Val | accuracy: 50.16 -> 51.42 | Improved |
| use token-wise query conditions instead of a global query condition (ablation) | HQGA · MSRVTT-QA Val | accuracy: 37.52 vs. 37.03 | Mixed |
Closest papers by meaning
- Language-aware Visual Semantic Distillation for Video Question Answering (CVPR 2024 · similarity 0.92)
- Commonsense Video Question Answering through Video-Grounded Entailment Tree Reasoning (CVPR 2025 · similarity 0.90)
- Divide and Conquer: Exploring Language-centric Tree Reasoning for Video Question-Answering (ICML 2025 · similarity 0.89)
- Not Another Text Benchmark: Putting the “Visual" Back in Visual Question Answering for Large Video Models (NeurIPS 2026 · similarity 0.89)
- Granularity-Adaptive Spatial Evidence Tokenization for Video Question Answering (AAAI 2025 · similarity 0.89)
- VideoEspresso: A Large-Scale Chain-of-Thought Dataset for Fine-Grained Video Reasoning via Core Frame Selection (CVPR 2025 · similarity 0.89)
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.
