List acceptence for scale parameter

[onurtore]

Hi!

I modified the coreml framework, which results with scale parameter in convert operation to accept a list, rather than a single float.

Here are my changes:

I can generalize this to different shapes just like bias parameter. Should I create a PR for this? Would be this helpful? Is there a specific reason why team has not implemented this functionality?

[TobyRoseman]

Making changes to the coremltools repository is not going to change the Core ML Framework.

What problem are you trying to solve?

[onurtore]

Making changes to the coremltools repository is not going to change the Core ML Framework.

What problem are you trying to solve?

I did not exactly understand your response. What is the difference between CoreML Framework and coremltools repository? Maybe I wrongly use the term CoreML Framework. What I want to achieve is standard operation for image processing. We have 3 channel images (RGB) and I want to have seperate scale parameter for each of the channels, similar to torchvision. I believe mul ops is supported but the _insert_image_preprocessing_ops is lacking the support.

Unfortunately, looks like we have communication issue, I did not understand you clearly, and I felt like what I am trying to say was clear which is obviously wrong. Can you eleborate more on your side. I hope we can understand each other.

[onurtore]

@TobyRoseman friendly ping.