为什么backend模型初始化的时候，CLASSES和PALETTE要从data type的配置获取，而不是从checkpoint加载。 #198
Answered
by
grimoire
liuyunfeng2016
asked this question in
Q&A
-
|
因为一些自定义的dataset type并没有确定这里的值，只在初始化之后，才获取到这两个类的值。这个是不是应该和初始化pytorch model保持一致，直接从checkpoint中读取更好 |
Beta Was this translation helpful? Give feedback.
Answered by
grimoire
Apr 14, 2022
Replies: 1 comment 1 reply
-
|
我向其他repo的开发者确认了一下，事实上现在的实现里并没有办法保证CLASSES或PALLETTE等属性一定存在于checkpoint中。 |
Beta Was this translation helpful? Give feedback.
1 reply
-
|
好吧，那就需要自定义的类，像官方一样，将全局变量的值的类别给定死，感谢你的回复 |
Beta Was this translation helpful? Give feedback.
Answer selected by
RunningLeon
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
我向其他repo的开发者确认了一下，事实上现在的实现里并没有办法保证CLASSES或PALLETTE等属性一定存在于checkpoint中。