Research map / Vision-language and multimodal / Multimodal LLMs
BI-MDRG: Bridging Image History in Multimodal Dialogue Response Generation
Hee Suk Yoon, Eunseop Yoon, Joshua Tian Jin Tee, Kang Zhang, Yu-Jung Heo, Du-Seong Chang, Chang D. Yoo
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 multimodal causal attention mask modulation (ablation) | BI-MDRG · ImageChat (zero-shot) | BLEU-1: 10.0 -> 10.9 | Improved |
| feed image features through cross-attention instead of text-only image history (BI-MDRG vs Divter_LLM) | Divter_LLM · ImageChat (zero-shot) | BLEU-1: 8.6 -> 10.9 | Improved |
| replace Divter_LLM with BI-MDRG on MMDialog textual image description | Divter_LLM · MMDialog | Textual image description BLEU-1: 44.2 -> 52.2 | Improved |
| replace Divter_LLM with BI-MDRG on MMDialog image quality | Divter_LLM · MMDialog | Image response IS: 21.0 -> 22.4 | Improved |
| BI-MDRG (4B) vs MiniGPT5 (9B) on text response BLEU-1 | BI-MDRG · MMDialog | Text response BLEU-1: 27.6 vs 29.1 | Hurt |
| BI-MDRG vs retrieval/classifier intent models on image intent F1 | BI-MDRG · PhotoChat | Intent F1: 55.7 vs 58.9 (T5-3B) | No clear effect |
| add Citation Module instead of LLMCite for citation tags (ablation) | BI-MDRG · MDIC | DINOv2 image consistency: 0.34 -> 0.53 | Improved |
| add LLMCite citation tags over no citations (4B) (ablation) | BI-MDRG 4B · MDIC | DINOv2 image consistency: 0.25 -> 0.34 | Improved |
| add citation tags with customized text-to-image (ablation) | BI-MDRG · MMDialog | Intent/TID/TR dialogue response metrics: TR B1 27.4 -> 27.6 | No clear effect |
| scale VLM size from 4B to 9B without citation framework (ablation) | BI-MDRG · MDIC | DINOv2 image consistency: 0.25 -> 0.26 | No clear effect |
| scale VLM size from 4B to 9B (ablation) | BI-MDRG · MMDialog | Text response BLEU-1: 27.4 -> 30.3 | Improved |
| replace LLMCite with trained citation prediction | Divter_LLM + LLMCite · MDIC | Citation token prediction accuracy: 33.5 -> 84.0 | Improved |
Later papers that compared with this method
1 later accepted paper reports BI-MDRG as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Making Visual Dialogue More Engaging: A New Task, Method, and Metric (AAAI 2026 · better on 8 of 8 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
- Structure-Aware Multimodal Sequential Learning for Visual Dialog (AAAI 2024 · similarity 0.89)
- F2RVLM: Boosting Fine-grained Fragment Retrieval for Multi-Modal Long-form Dialogue with Vision Language Model (AAAI 2026 · similarity 0.88)
- MMDU: A Multi-Turn Multi-Image Dialog Understanding Benchmark and Instruction-Tuning Dataset for LVLMs (NeurIPS 2024 · similarity 0.87)
- V^2Dial: Unification of Video and Visual Dialog via Multimodal Experts (CVPR 2025 · similarity 0.87)
- WEAVE: Unleashing and Benchmarking the In-context Interleaved Comprehension and Generation (CVPR 2026 · similarity 0.86)
- SEED-Bench: Benchmarking Multimodal Large Language Models (CVPR 2024 · similarity 0.86)
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.
