关于原模型CLIPscore的一些咨询
#335
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
{{ message }}
-
您好，我利用该工程文件中的evaluate.py脚本对Taiyi-Stable-Diffusion-1B-Chinese-v0.1模型进行评估，使用Taiyi-CLIP-RoBERTa-102M-ViT-L-Chinese的文本编码器，open_clip的ViT-L-14的图像编码器，在Flickr30k-CNA的test数据集上跑了1000个样本，最后评估出来只有'avg_clip': 0.16480967822718667，请问这个结果是合理的吗？
感谢您的解答
Beta Was this translation helpful? Give feedback.
All reactions