search

xiangning-chen / SmoothDARTS

Code for our ICML'2020 paper "Stabilizing Differentiable Architecture Search via Perturbation-based Regularization"
76 stars 12 forks source link

ImageNet training details #8

Closed Minione closed 3 years ago

Minione commented 3 years ago

Dear authors,

Thanks for providing the code of SDARTS.

I noticed that in sota/cnn/train_imagenet.py the batch_size is set to 1024. Does that mean the ImageNet results of SDARTS in the paper were obtained with 8-GPU training, i.e., the same setup used by PC-DARTS / P-DARTS?

Thanks!

Minione commented 3 years ago

BTW, if you still have the information, could you kindly provide the cell designs found by SDARTS on CIFAR-10 from the original DARTS space? I only found cells found from S1-4 space in sota/cnn/genotypes.py but not those found from the DARTS space. Thanks!

xiangning-chen commented 3 years ago

Dear authors,

Thanks for providing the code of SDARTS.

I noticed that in sota/cnn/train_imagenet.py the batch_size is set to 1024. Does that mean the ImageNet results of SDARTS in the paper were obtained with 8-GPU training, i.e., the same setup used by PC-DARTS / P-DARTS?

Thanks!

Thanks for your attention. Yes, we use 8-GPU to train our models on ImageNet.

xiangning-chen commented 3 years ago

BTW, if you still have the information, could you kindly provide the cell designs found by SDARTS on CIFAR-10 from the original DARTS space? I only found cells found from S1-4 space in sota/cnn/genotypes.py but not those found from the DARTS space. Thanks!

The DARTS space is S5, so you can find the searched cells here: https://github.com/xiangning-chen/SmoothDARTS/blob/master/sota/cnn/genotypes.py#L26

Minione commented 3 years ago

Thanks for the prompt reply! That's super helpful!