search

idstcv / ZenNAS

218 stars 35 forks source link

GENet how to use soft-labels #24

Closed bigcash closed 1 year ago

bigcash commented 1 year ago

i follow the paper "Neural Architecture Design for GPU-Efficient Networks", and see "We also use ResNet-152 as teacher network to get soft-labels. The soft-label loss and the true-label loss are weighted 1:1." at page 8 how to use soft-labels，teach genet. like knowledge distillation? have any code in this repo?

dovedx commented 1 year ago

这是来自QQ邮箱的假期自动回复邮件。   邮件我已经成功接收，我会尽快处理并答复，谢谢！

MingLin-home commented 1 year ago

Hi Bigcash, The code for knowledge distillation is here: [https://github.com/idstcv/ZenNAS/blob/main/ts_train_image_classification.py]()

qq272574497 commented 1 year ago

Is it only suitable for VCNN network? Does the residual network fit?

dovedx commented 1 year ago

这是来自QQ邮箱的假期自动回复邮件。   邮件我已经成功接收，我会尽快处理并答复，谢谢！

qq272574497 commented 1 year ago

Can I communicate with you via email

MingLin-home commented 1 year ago

Can I communicate with you via email

Sure, why not? We are open to all kinds of discussions and collaborations. Please send me email for something private. I will receive the notification email as well if you reply in this thread. So no difference except for privacy.