How and when can future versions of the custom deep learning layer improve flexibility?

I found that the custom deep learning layer is very tasteless and can only be used for simple image classification or regression. It is difficult to customize the input and output freely. This problem has been raised since a year ago, but the MathWorks company has not resolved this aspect. May I give up in the future Is it this way?

Answer 1

[Steven Lord]

If you wish to request that the custom deep learning layer functionality be enhanced with specific functionality (like the ability to "customize the input and output freely" -- what specifically do you mean by 'customize'?) you can file enhancement requests by contacting Technical Support using the Contact Support link on the Support section of this site.

[cui,xingxing]

Specific reference here:

'customize', https://www.mathworks.com/matlabcentral/answers/1427-what-frustrates-you-about-matlab#comment_838871

https://www.mathworks.com/matlabcentral/answers/1427-what-frustrates-you-about-matlab#answer_464291

matlab-missing here:

https://www.mathworks.com/matlabcentral/answers/1325-what-is-missing-from-matlab#answer_470158

related issue:

https://www.mathworks.com/matlabcentral/answers/524111-arcface-loss-sphereface-loss?s_tid=srchtitle