Michael Tschannen
Results
11
comments of
Michael Tschannen
You can fine-tune with the sigmoid loss/SigLIP training code using [this config](https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/image_text/siglip_lit_coco.py). You'll have to slightly adapt the model code (variant, pool_type, init with the pretrained weights), preprocessing function, schedule...