Regnet Results

[andravin]

Why is the regnety_032 result (82.008% top-1) better than all others, including regnety_320 (80.798%)?

Was it trained differently from the others?

Where can I find the training hyper-parameters for the Regnet results?

https://github.com/rwightman/pytorch-image-models/blob/master/results/results-imagenet.csv

More generally, where can I find training hyper-parameters for all pytorch-image-models results?

[andravin]

I learned that regenety_032 result was trained by Ross while the others were not. So it was probably trained much longer and with stronger data augmentation.

[rwightman]

Yup, lots of aug, longer training, more regularization ... I added stochastic depth to the model.