What is the training details on CUB-200？

recasens / Saliency-Sampler

The saliency-based is a distortion layer for convolutional neural networks that helps to improve the spatial sampling of input data for a given task.

163 stars 23 forks source link

What is the training details on CUB-200？ #2

Closed DolphinAC closed 5 years ago

DolphinAC commented 5 years ago

I find the train loss can not descend on CUB-200. Could you tell me the training details ,such as learning rate on taskNet and saliencyNet， pretrained dataset，epochs and so on.