SigLIP

定义

SigLIP (Sigmoid Loss for Language Image Pre-Training) is a vision-language pre-training method that replaces softmax loss with sigmoid loss for more efficient training.