New framework enables zero-shot captioning of Indonesian traditional clothing

Researchers have developed Custom ZeroCLIP, a novel retrieval-augmented vision-language framework designed for the zero-shot captioning of traditional Indonesian clothing. This framework utilizes a combination of CLIP and BERT text encoders with an LSTM decoder, trained on a dataset of 3,800 expert-annotated images. By employing a province-level inductive zero-shot protocol, the model demonstrates strong performance on unseen provinces, achieving a CLIPScore of 0.8536 and outperforming existing baselines. AI

IMPACT This research advances zero-shot learning capabilities for specialized cultural heritage datasets, potentially improving AI's ability to understand and describe diverse cultural artifacts.

RANK_REASON The cluster describes a research paper published on arXiv detailing a new framework for image analysis. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• BERT
• Custom ZeroCLIP
• LSTM

• paper
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →