Research map / Video / Video-language understanding
Contrastive Video Question Answering via Video Graph Transformer
Junbin Xiao, Pan Zhou, Angela Yao, Yicong Li, Richang Hong, Shuicheng Yan, 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 dynamic graph transformer (DGT) instead of mean pooling (ablation) | VGT · TGIF-QA Action | accuracy: 89.6 -> 95.0 | Improved |
| add node transformer (ablation) | VGT · NExT-QA val | Acc@All: 53.84 -> 55.02 | Improved |
| add edge transformer (ablation) | VGT · NExT-QA val | Acc@All: 54.30 -> 55.02 | Improved |
| add temporal graph transformer (ablation) | VGT · STAR Val | Acc@M: 42.37 -> 44.27 | Improved |
| add frame-level feature F_I to object graphs (ablation) | VGT · NExT-QA val | Acc@All: 53.58 -> 55.02 | Improved |
| contrastive answer matching instead of [CLS] classification (ablation) | VGT · NExT-QA | Acc@All: 45.82 -> 55.02 | Improved |
| add self-supervised contrastive loss L_vq (ablation) | VGT (BERT) · NExT-QA | Acc@All: 55.02 -> 57.15 | Improved |
| add self-supervised contrastive loss L_vq (ablation) | VGT (BERT) · TGIF-FrameQA | accuracy: 61.6 -> 61.3 | No clear effect |
| replace BERT with RoBERTa (no L_vq) (ablation) | VGT · NExT-QA / STAR-QA | Acc@All / Acc@M: 55.02 -> 58.45 (NExT-QA); 44.27 -> 42.44 (STAR-QA) | Mixed |
| add 4 negative questions in L_vq (ablation) | CoVGT · NExT-QA | accuracy: 58.45 -> 60.01 | Improved |
| hard negatives by question type instead of random (ablation) | CoVGT · NExT-QA | accuracy: 59.35 -> 60.01 | Improved |
| pretrain on 0.18M WebVid data (ablation) | VGT · NExT-QA val | Acc@All: 55.02 -> 56.02 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- MERLOT — MERLOT: Multimodal Neural Script Knowledge Models (NeurIPS 2021) · better on 1 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Video Graph Transformer for Video Question Answering (ECCV 2022 · similarity 0.95)
- Video-Context Aligned Transformer for Video Question Answering (AAAI 2024 · similarity 0.91)
- 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 0.89)
- MoReVQA: Exploring Modular Reasoning Models for Video Question Answering (CVPR 2024 · similarity 0.89)
- Language-aware Visual Semantic Distillation for Video Question Answering (CVPR 2024 · 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.
