CNN Regression Image to Vector

Hello,

I am trying to train a CNN network to take a large group of 128X128X1 .png images and learn how to predict a 5X1 vector of elements. I have matching pairs of the data but too much to read into memory like the digit rotation example does. Can anyone point me to an example of how to create a datastore out of data like this? I am also unclear on how to use a datastore to train a regression network since I don't have xTrain and yTrain readily available like the example does. Any help would be appriciated. Thanks,

--Joe

Answer 1

[Srivardhan Gadila]

Refer to Image Regression using .mat Files and a datastore MATLAB Answer and following other links:

1. Is there a regression layer for Convolutional neural network toolbox?
2. Train Network Using Out-of-Memory Sequence Data

[Joseph Galbraith]

Thanks Srivardhan! I think link #2 answered my question. I am still a little unsure about padding the target vector but I can figure it out. Thanks for your help.

[tianliang wang]

Hi,Gadila, if i want to using a image datastore to do the image to vector regression, what should i do ? thank you!