Research map / Vision-language and multimodal / Multimodal LLMs
Aligning Vision to Language: Annotation-Free Multimodal Knowledge Graph Construction for Enhanced LLMs Reasoning
Junming Liu, Siyuan Meng, Yanting Gao, Song Mao, Pinlong Cai, Guohang Yan, Yirong Chen, Zilin Bian, Ding Wang, Botian Shi
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 VaLiK image-only knowledge graph to Qwen2.5-7B | Qwen2.5-7B · CrisisMMD | accuracy (avg across tasks): +4.41% | Improved |
| add VaLiK text-image knowledge graph to Qwen2.5-7B | Qwen2.5-7B · CrisisMMD | accuracy (avg across tasks): +4.90% | Improved |
| add VaLiK graph to larger Qwen2.5-32B/72B | Qwen2.5-32B/72B · CrisisMMD | accuracy: +2.0% to 2.5% | Improved |
| add VaLiK KG to VLMs at inference | LLaVA-34B · CrisisMMD | accuracy: +2.41% image-only, +3.59% text-image | Improved |
| add CoE-based VLM description generation (ablation) | VaLiK · ScienceQA average (Text-Image KG) | accuracy (%): 74.72 -> 80.57 | Improved |
| add cross-modal similarity verification pruning (ablation) | VaLiK · CrisisMMD Task 2 (Image-only KG) | accuracy (%): 47.00 -> 49.54 | Improved |
| add similarity verification on ScienceQA image-only natural science (ablation) | VaLiK · ScienceQA NAT (Image-only KG) | accuracy (%): 80.06 -> 79.14 | Hurt |
| add similarity verification on ScienceQA overall average (Text-Image KG) (ablation) | VaLiK · ScienceQA average (Text-Image KG) | accuracy (%): 80.57 -> 83.16 | Improved |
| use VaLiK MMKG instead of pre-built Visual Genome or Mmkg | Qwen2.5-7B · ScienceQA average | accuracy (%): 75.08 (VG) -> 83.16 | Improved |
| use Mmkg as external knowledge for Qwen2.5-7B | Qwen2.5-7B · ScienceQA average | accuracy (%): 74.72 -> 73.47 | Hurt |
| add more VLMs to the CoE cascade (ablation) | VaLiK · CrisisMMD | accuracy | No clear effect |
| use larger retrieval model (32B/72B) for LightRAG retrieval (ablation) | VaLiK · retrieval | performance: 1%-5% improvement | Improved |
Closest papers by meaning
- Mario: Multimodal Graph Reasoning with Large Language Models (CVPR 2026 · similarity 0.91)
- UKnow: A Unified Knowledge Protocol with Multimodal Knowledge Graph Datasets for Reasoning and Vision-Language Pre-Training (NeurIPS 2024 · similarity 0.90)
- LoMo: Local Modality Substitution for Deeper Vision-Language Fusion (NeurIPS 2026 · similarity 0.90)
- Insight-V: Exploring Long-Chain Visual Reasoning with Multimodal Large Language Models (CVPR 2025 · similarity 0.90)
- GraphVLM: Benchmarking Vision Language Models for Multimodal Graph Learning (CVPR 2026 · similarity 0.89)
- Analyzing Fine-Grained Alignment and Enhancing Vision Understanding in Multimodal Language Models (NeurIPS 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.
