fashion-clip

Why we included this project

FashionCLIP is a CLIP-style vision-language model fine-tuned on fashion data, so it can match product images to free-text descriptions and run zero-shot classification without per-category training. That makes it a practical starting point for e-commerce search, product tagging, or recommendation features. The repo ships the fine-tuned weights on Hugging Face, Colab notebooks covering feature extraction, classification, and evaluation with RecList, and a Streamlit demo for interactive testing. The README also documents the switch to a LAION-based checkpoint (FashionCLIP 2.0) with benchmark numbers, which helps you judge whether the upgrade is worth it. For a small team that wants a domain-tuned multimodal backbone instead of training one from scratch, this is a low-friction option.