search

tobran / GALIP

[CVPR2023] A faster, smaller, and better text-to-image model for large-scale training
MIT License
225 stars 25 forks source link

你好，我想请问一下辨别器的CLIP_feat是怎么获得的 #26

Open LLLLLLiuL opened 4 months ago

LLLLLLiuL commented 4 months ago

image 我在这里有看到CLIP_feat设置为true之后，在D_BLOCK中return x + self.gamma res + self.beta CLIP_feat， 请问CLIP_feat这个参数是怎么获得的呀，感谢！