Join 4 outputs from different CNN into 1 output

Hello there, I have the following architecture:

I train 4 differents CNN, with 4 different scales and N randoms crops of each. I have 8 classes in total. After training I have each CNN trained with 4 final outputs (4 outputs with 8 classes each). The 8 classes are the same. But when I have to validate I want to have only 1 output with 8 classes. I dont know how to join the 4 outputs (FC layers) into output (FC layer) of 8 classes, like in the picture.

The CNNs are ResNet50.

Thank you very much.

If you are getting 4 class results (max_index) out of the 4 classifiers, just take majority voting. That would be the simplest way to ensemble the results.

Thanks @Anshumaan_Dash I’ll try