Research map / Vision-language and multimodal / Visual question answering
New Datasets and Models for Contextual Reasoning in Visual Dialog
Yifeng Zhang, Ming Jiang, Qi Zhao
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 memory mechanism to NDM (VQA) (ablation) | NDM · CLEVR-VD / GQA-VD | accuracy: 55.27 -> 57.81 (CLEVR-VD); 54.95 -> 57.63 (GQA-VD) | Improved |
| add the contextual modules (Include/Exclude) to NDM (VQA) (ablation) | NDM · CLEVR-VD / GQA-VD | accuracy: 55.27 -> 58.12 (CLEVR-VD); 54.95 -> 57.98 (GQA-VD) | Improved |
| add both memory and contextual modules to NDM (VQA) (ablation) | NDM · CLEVR-VD / GQA-VD | accuracy: 55.27 -> 59.80 (CLEVR-VD); 54.95 -> 60.84 (GQA-VD) | Improved |
| add memory and contextual modules to CorefNMN (ablation) | CorefNMN · CLEVR-VD / GQA-VD | accuracy: 54.69 -> 57.72 (CLEVR-VD); 54.06 -> 58.87 (GQA-VD) | Improved |
| add the memory mechanism alone to CorefNMN (ablation) | CorefNMN · CLEVR-VD / GQA-VD | accuracy: 54.69 -> 56.45 (CLEVR-VD); 54.06 -> 56.24 (GQA-VD) | Improved |
| add the contextual modules alone to CorefNMN (ablation) | CorefNMN · CLEVR-VD / GQA-VD | accuracy: 54.69 -> 56.71 (CLEVR-VD); 54.06 -> 56.46 (GQA-VD) | Improved |
| add pretrained ViLBERT features to NDM | NDM · CLEVR-Dialog / CLEVR-VD / VisDial / GQA-VD | accuracy: 68.21 -> 68.23 (CLEVR-Dialog); 59.89 -> 59.92 (CLEVR-VD); 52.72 -> 52.91 (VisDial); 60.84 -> 61.08 (GQA-VD) | No clear effect |
| use NDM instead of CorefNMN | CorefNMN · GQA-VD | accuracy: 56.59 -> 60.84 | Improved |
| use NDM instead of VD-BERT | VD-BERT · VisDial | accuracy: 51.63 -> 52.72 (VisDial); 59.67 -> 59.89 (CLEVR-VD) | Mixed |
| increase number of contextual dependencies of a question from 0 to 3 or more | NDM · GQA-VD | accuracy: 61.28 -> 59.92 | No clear effect |
| increase number of contextual dependencies of a question from 0 to 3 or more | BUTD · CLEVR-VD | accuracy: 60.95 -> 47.82 | Hurt |
| increase number of contextual dependencies of a question from 0 to 3 or more | CorefNMN · CLEVR-VD | accuracy: 60.83 -> 53.74 | Hurt |
Closest papers by meaning
- Visual Graph Arena: Evaluating Visual Conceptualization of Vision and Multimodal Large Language Models (ICML 2025 · similarity 0.88)
- ConTextual: Evaluating Context-Sensitive Text-Rich Visual Reasoning in Large Multimodal Models (ICML 2024 · similarity 0.88)
- VGR: Visual Grounded Reasoning (ICLR 2026 · similarity 0.88)
- Structure-Aware Multimodal Sequential Learning for Visual Dialog (AAAI 2024 · similarity 0.88)
- Re-ranking Reasoning Context with Tree Search Makes Large Vision-Language Models Stronger (ICML 2025 · similarity 0.88)
- Prism: A Framework for Decoupling and Assessing the Capabilities of VLMs (NeurIPS 2024 · similarity 0.88)
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.
