Training option ExecutionEnvironment = "parallel" does not work with trainMaskRCNN.

I am trying to use trainMaskRCNN to train a maskrcnn model. In the training options, I have set ExecutionEnvironment = "parallel", but this results in the error "Error using minibatchqueue. Expected OutputEnvironment to match one of these values: 'auto', 'gpu', 'cpu'".

The problem is that trainMaskRCNN passes the ExecutionEnvironment value to the function minibatchqueue, which only accepts one of the three values above. How are you then supposed to be able to use "parallel" with trianMaskRCNN?

Answer 1

[T.Nikhil kumar]

Hello David,

I understand that you want to set the ‘ExecutionEnvironment’ to ‘parallel’ for training your Mask RCNN model but are facing an error about unsupported environment.

Support for ‘multi-gpu’ and ‘parallel’ options for the ‘ExectionEnvironment’ property has been added for the MaskRCNN training from the MATLAB R2023b release. I would suggest you update MATLAB to R2023b release or later to be able to use ‘parallel’ execution environment for ‘trainMaskRCNN’.

Refer to the ‘ExecutionEnvironment’ property in the following documentation to know about all supported options:

https://www.mathworks.com/help/deeplearning/ref/trainingoptions.html

Hope this helps you proceed further!