patrickjohncyh/fashion-clip

Built on contrastive vision-language learning, FashionCLIP fine-tunes LAION CLIP checkpoints on 700K+ fashion image-text pairs from Farfetch, enabling zero-shot performance on domain-specific tasks like retrieval, classification, and attribute parsing. The model integrates with Hugging Face's transformers library and provides a dedicated Python API for efficient batch encoding of images and text into aligned embedding spaces, achieving significantly higher accuracy (F1 scores 0.62–0.83) across fashion benchmarks compared to general-purpose CLIP variants.

497 stars. No commits in the last 6 months.