How to combine multiple inputs and a single output in a datastore?

Hi,

As I know, a function of combine can make a datastore containing several datastores. But here, in order to use "trainnetwork" for deep learning, we have to use only one datastore. How can I seperate input datastores to the output one to train my network? I guess labels of input parts may have to be created.

I will use two inputs (images + numerical data) and one output (images) with unet which is already built in Matlab.

[Mahesh Taparia]

Hi,

Is the size of numerical data same as input images?

If it is so, you can concatenate the numerical data with the input data in the third dimension (i.e. depth) and train your network.

[taicheng chen]

same problem QQ

In my problem, the input images size and numerical data size are different.

[Sunusi Abdullahi]

you can use the function of data augment first, to make the size of your inputs the same, then you can concatenate

Answer 1

[Jayanti]

Hi Sungkun,

I found a similar Matlab answer related to your query. This answer explains how to create and combine datastores for multiple inputs.

https://www.mathworks.com/matlabcentral/answers/586949-multi-input-imagedatastore

You can use “imageDatastore” for input and output images. For numerical data, you can use “arrayDatastore” or “tabularTextDatastore” depending on the format of your data.

I am also attaching the documentation link on various datastores for your reference.

https://www.mathworks.com/help/matlab/datastore.html

Then use the “combine” function to combine multiple datastores into single datastores.

Hope it will be helpful!