search

GAIA-vision / GAIA-ssl

Apache License 2.0
18 stars 3 forks source link

Training details #2

Closed NEUdeep closed 2 years ago

NEUdeep commented 2 years ago

good job.

I have some questions about Training details.

  1. In your paper, “We train 200 epochs using a batch size of 1024 on 16 GPUs and an initial learning rate of 0.12”, which type of GPUs are used in your paper? V100?
  2. you compute the Spearman [36] correlation between the similarity ranking and their final performance ranking. Why not use Kendall’s Tau correlation?
  3. Is there any data on search costs?

thanks

NickChang97 commented 2 years ago

1) Yes, we trained supernet with V100. 2) We follow UnNAS, so we adapt the Spearman. And, Kendall's Tau is truly a more strict judgment. 3) I don't understand this question, can you explain it more ?