Research map / Video / Video-language understanding
Hierarchical Conditional Relation Networks for Multimodal Video Question Answering
Thao Minh Le, Vuong Le, Svetha Venkatesh, Truyen Tran
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 motion conditioning (ablation) | HCRN · TGIF-QA Action | accuracy: 69.8 -> 75.2 | Improved |
| add linguistic (question) conditioning (ablation) | HCRN · TGIF-QA Action | accuracy: 68.7 -> 75.2 | Improved |
| add relations (k_max=n-1, t=3) over no relations (k_max=1,t=1) (ablation) | HCRN · TGIF-QA Action | accuracy: 72.8 -> 75.2 | Improved |
| add relations (k_max=n-1) over no relations (ablation) | HCRN · TGIF-QA Count | MSE: 3.87 -> 4.00 | Hurt |
| add hierarchy (two levels) over video-level CRN only (ablation) | HCRN · TGIF-QA Action | accuracy: 72.7 -> 75.2 | Improved |
| add hierarchy (two levels) over video-level CRN only (ablation) | HCRN · TGIF-QA FrameQA | accuracy: 57.2 -> 55.9 | Hurt |
| add multiplicative relation in all CRNs (ablation) | HCRN · TGIF-QA FrameQA | accuracy: 74.0 -> 75.2 Action; 81.7 -> 81.3 Transition | Mixed |
| sample subsets directly instead of from pre-computed superset (ablation) | HCRN · TGIF-QA Action | accuracy: 75.2 -> 75.3 | No clear effect |
| add subtitle pre-selection to S.+Q. baseline (ablation) | (B) S.+Q. · TVQA val (w/o timestamp) | accuracy: 42.3 -> 57.9 | Improved |
| add subtitle pre-selection to S.+Q. baseline (ablation) | (B) S.+Q. · TVQA val (w/ timestamp) | accuracy: 41.8 -> 62.1 | Improved |
| replace baseline with HCRN CRN units in textual stream | (B) S.+Q. w/ pre-selection · TVQA val (w/o timestamp) | accuracy: 57.9 -> 59.0 | Improved |
| replace averaged visual features with HCRN visual stream | (B) V.+Q. · TVQA val | accuracy: 42.2 -> 43.1 | Improved |
Closest papers by meaning
- Hierarchical Conditional Relation Networks for Video Question Answering (CVPR 2020 · similarity 0.94)
- Video as Conditional Graph Hierarchy for Multi-Granular Question Answering (AAAI 2022 · similarity 0.92)
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.89)
- Cross-modal Causal Relation Alignment for Video Question Grounding (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 0.89)
- Modularized Self-Reflected Video Reasoner for Multimodal LLM with Application to Video Question Answering (ICML 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.
