Jia Guo
Jia Guo
Use normed_embedding instead
先用onnx_helper.py自己测一下看看
how about the test accuracy?
In my experience, if we train for numerous epochs, the scale of feature vector will decrease.
I think if the accuracy increase, the scale doesn't matter.
use the embedding after l2-normalization
模型就是指backbone.
The swapping demo had been moved to discord bot.
The 128 model does not perform as well as the Discord version. At present, there is no reason to upload it again.
> But seems the discord app is down at the moment No, it's still in service. > BTW, is the discord version swap model will be public in the future?...