Preparation of the dataset about densenas HOT 2 CLOSED

Thanks for your quick reply. I notice that so I close the issue.

from densenas.

Hi @colorjam !
Thanks for your interest in our work. The training data is organized as the standard ImageNet, i.e.,
-ImageNet
--train
---nxxx
--val
---nxxx
The non-folder items will be skipped by if not os.path.isdir(split_path).

DenseNAS/dataset/mk_split_img_list.py

Line 8 in 883bc04

split_path = os.path.join(data_path, 'train_img')

I name the origin train folder as train_img, and this may cause some issues in your experiment. I have changed it to train.
Thanks a lot for reminding me.

from densenas.