Research map / Vision-language and multimodal / Multimodal LLMs
CiQi-Agent: Aligning Vision, Tools and Aesthetics in Multimodal Agent for Cultural Reasoning on Chinese Porcelains
Wenhan Wang, Zhixiang Zhou, Zhongtian Ma, Yanzhu Chen, Ziyu Lin, Hao Sheng, Pengfei Liu, Wenqi Shao, Qiaosheng Zhang, Yu Qiao
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 SFT on CiQi-VQA (ablation) | Qwen2.5-VL-7B-Instruct · CiQi-Bench (multiple-choice) | average accuracy: 67.4 -> 75.9 | Improved |
| add SFT on CiQi-VQA (ablation) | Qwen2.5-VL-7B-Instruct · CiQi-Bench (multiple-choice) | naming accuracy: 69.0 -> 81.9 | Improved |
| add SFT on CiQi-VQA (ablation) | Qwen2.5-VL-7B-Instruct · CiQi-Bench (free-form) | average accuracy: 36.4 -> 59.1 | Improved |
| add SFT on CiQi-VQA (ablation) | Qwen2.5-VL-7B-Instruct · CiQi-Bench (free-form) | dynasty accuracy: 20.3 -> 64.6 | Improved |
| add SFT on CiQi-VQA (ablation) | Qwen2.5-VL-7B-Instruct · CiQi-Bench (multiple-choice) | reign period accuracy: 60.4 -> 53.5 | Hurt |
| add tool-aware RL after SFT (ablation) | CiQi-Agent · CiQi-Bench (multiple-choice) | average accuracy: 69.7 -> 81.5 | Improved |
| add tool-aware RL after SFT (ablation) | CiQi-Agent · CiQi-Bench (multiple-choice) | kiln accuracy: 67.2 -> 81.8 | Improved |
| add tool-aware RL after SFT (ablation) | CiQi-Agent · CiQi-Bench (free-form) | average accuracy: 62.3 -> 66.7 | Improved |
| add vision tool to base model (ablation) | GPT-5 · CiQi-Bench (multiple-choice) | average accuracy: 75.8 -> 76.3 | No clear effect |
| add retrieval tools to base model (ablation) | GPT-5 · CiQi-Bench (multiple-choice) | average accuracy: 75.8 -> 70.8 | Hurt |
| add all tools to base model (ablation) | Qwen2.5-VL-7B-Instruct · CiQi-Bench (multiple-choice) | average accuracy: 67.4 -> 48.1 | Hurt |
| add vision tool (ablation) | CiQi-Agent · CiQi-Bench (multiple-choice) | average accuracy: 69.7 -> 75.1 | Improved |
Closest papers by meaning
- VaseVQA-3D: Benchmarking 3D VLMs on Ancient Greek Pottery (ICLR 2026 · similarity 0.83)
- Unlocking the Essence of Beauty: Advanced Aesthetic Reasoning with Relative-Absolute Policy Optimization (ICLR 2026 · similarity 0.81)
- EpiAgent: An Agent-Centric System for Ancient Inscription Restoration (CVPR 2026 · similarity 0.80)
- Beyond Aesthetics: Cultural Competence in Text-to-Image Models (NeurIPS 2024 · similarity 0.80)
- Bridging Cognitive Gap: Hierarchical Description Learning for Artistic Image Aesthetics Assessment (AAAI 2026 · similarity 0.80)
- Component-Level Segmentation for Oracle Bone Inscription Decipherment (AAAI 2025 · similarity 0.80)
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.
