BROWSE
Detail
Conference Paper
ChemCLIP: 화학 성분 정보를 활용한 다중 모달리티 기반 문화유산 속성 추론 연구
Cited - time in 
Share 
Share
- Issue Date
- 2026-06
- Citation
- 대한전자공학회 학술 대회 (하계) 2026, pp.1-5
- Publisher
- 대한전자공학회
- Language
- Korean
- Type
- Conference Paper
- Abstract
- This paper proposes ChemCLIP as a multimodal learning framework for cultural heritage attribute inference integrating visual, textual, and chemical composition information. ChemCLIP encodes images, attribute texts, and chemical compositions using modality-specific encoders and aligns them in a shared embedding space. Attribute inference is performed by similarity matching between input representations and candidate attribute texts without task-specific classifiers. Experiments on a cultural heritage dataset of 250 samples validate the effectiveness of ChemCLIP with 69.4% text attribute retrieval accuracy and 88.7% composition type classification accuracy.
- KSP Keywords
- Attribute inference, Composition Information(CI), Cultural Heritage, Embedding space, Learning framework, Multimodal learning, Retrieval accuracy, Similarity Matching, Task-specific, Type classification, chemical composition