How to create multiple image embeddings from a single text embedding

[tenten0727]

I wanted to generate a variety of image embeddings for a single text input, but when I ran the code, I found that it generated the same image embeddings for each text input.

To generate a variety of image embeddings from a single text input, I changed line 286 of model_lib/decoder/clip_prior.py to

self.std = self.log_std_init.exp()

Is this collect?

Thank you very much!