Research map / Vision-language and multimodal / Visual question answering
Modality-Balanced Models for Visual Dialogue
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Combine image-only and joint models by test-time logit ensemble | Image-only and joint models · VisDial v1.0 val | NDCG: 61.20 vs 61.04 (Img-Only) and 58.97 (Joint) | Helped |
| Combine image-only and joint logits with consensus dropout fusion | Image-only and joint models · VisDial v1.0 val | NDCG: 58.97 -> 59.93 | Helped |
| Use image-only instead of joint context (ablation) | Visual Dialog model · VisDial v1.0 val | NDCG: 58.97 -> 61.04 | Helped |
| Use image-only instead of joint context (ablation) | Visual Dialog model · VisDial v1.0 val | MRR: 64.57 -> 61.25 | Hurt |
| Reduce joint-model history to 5 features (ablation) | Joint model · VisDial v1.0 val | NDCG: 57.81 -> 58.24 | Helped |
| Reduce joint-model history to 1 feature (ablation) | Joint model · VisDial v1.0 val | NDCG: 57.81 -> 59.29 | Helped |
| Reduce joint-model history to 1 feature (ablation) | Joint model · VisDial v1.0 val | MRR: 64.47 -> 62.86 | Hurt |
| Add round dropout to the cross-attention model (ablation) | CA · VisDial v1.0 val | NDCG: 57.81 -> 58.97 | Helped |
| Increase CDF instance dropout rate (ablation) | CDF · VisDial v1.0 val | NDCG: 59.40 (p=0.00) -> 60.11 (p=0.35) | Helped |
| Increase CDF instance dropout rate (ablation) | CDF · VisDial v1.0 val | MRR: 64.61 (p=0.00) -> 64.21 (p=0.35) | Hurt |
| Ensemble image-only and joint models rather than image-only models only (ablation) | Three-model ensemble · VisDial v1.0 val | MRR: 62.24 -> 65.04 | Helped |
| Ensemble image-only and joint models rather than joint models only (ablation) | Three-model ensemble · VisDial v1.0 val | NDCG: 59.84 -> 61.50 | Helped |
Closest papers by meaning
- Structure-Aware Multimodal Sequential Learning for Visual Dialog (AAAI 2024 · similarity 0.91)
- V^2Dial: Unification of Video and Visual Dialog via Multimodal Experts (CVPR 2025 · similarity 0.91)
- Cross-modal Information Flow in Multimodal Large Language Models (CVPR 2025 · similarity 0.90)
- MMDU: A Multi-Turn Multi-Image Dialog Understanding Benchmark and Instruction-Tuning Dataset for LVLMs (NeurIPS 2024 · similarity 0.89)
- F2RVLM: Boosting Fine-grained Fragment Retrieval for Multi-Modal Long-form Dialogue with Vision Language Model (AAAI 2026 · similarity 0.89)
- Understanding Information Storage and Transfer in Multi-Modal Large Language Models (NeurIPS 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.