Recently, I start Custom Vision from Azure, and I'm worried how to augment training images. I came up with data augment with affine transformation, but simultaneously guessed Custom Vision algorithm might include this process itself. Could you please tell me it is efficient to do data augmentation with affine transformation on Custom Vision from Azure?
Asked
Active
Viewed 98 times
0
-
which kind of affine transformation are you trying to do? mixing images and labels? – Ping Jin Dec 04 '19 at 19:52
-
Since we dont know which algorithm is used I think it can be dangerous. Azure probably use different algorithm. When doing data augmentation for a convolutional neural network can be usefull, doing data augmentation using manual feature extraction (e.g edge ) can destroy model performances. I would says custom vision does the augmentation if it's consider usefull to do it... But there is no true answer and since model change depending on data set, you should try with and without data augmentation every time you got new data – akhetos Dec 20 '19 at 12:21