senwu / dauphin Goto Github PK

Hi, How do we train imagenet data?

Hi. Interesting results with CIFAR-10. However, it is hard to deal with the larger dataset. The object "Causing the issue". I'm unable to use full dataset (74K training, 25K validation). How can we solve this?

Hello, I have been recently looking at your code for data augmentation on images, and I couldn't help but notice that aug_dataloaders on line 36 of image.py is initialized, but nothing is added into the list. It is passed to the EmmentalLearner along with the dataloaders, but why is this? I am curious because I am mainly focused on selecting the chosen transformations for data augmentation, but can't seem to grasp where I can do so.

Hi, I try the nice code and it is easy to run. But I did not get the exact numbers reported in the paper at the moment. Specifically, I got 97.72 for multi 4 loss 1 setting, while the reported number is 97.89. For multi 8 loss 4 setting, I got 98.10 while the reported one is 98.16. I would like to know whether I miss something important or it is just the variance. Thx