SigLIP 2 Style naflex

This is a finetune of the vision model of google/siglip2-base-patch16-naflex. It has been finetuned on over 90k styles with 4-5 images per style using an arcface loss with patch lengths of (256, 576, 784, 1024). The arcface head has been removed. The pooled embedding should now function as a style embedding.

Downloads last month
44
Safetensors
Model size
92.9M params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support