--scale_values and --mean_values of mo.py

[timosy]

When I quantize a model (and also train a model), I used scaling 

g_normalize = torchvision.transforms.Normalize(
      mean=[0.5029, 0.4375, 0.3465],
      std=[0.2818, 0.2659, 0.2629]

)
train_dataset = torchvision.datasets.ImageFolder(
      "/datasetA/train/",
      torchvision.transforms.Compose([
            torchvision.transforms.RandomResizedCrop(224), 
            torchvision.transforms.RandomHorizontalFlip(), 
            torchvision.transforms.ToTensor(),
            g_normalize
]))

In this case, do I also have to use the same values with "--scale_values" and "--mean_values" option when I convert the model to IR. or, it's not necessary to add these options?

[IntelSupport · Accepted solution]

Hi Hep77to,

Thanks for reaching out.

It is not necessary if the input pre-processing operations are part of the model. In this case, the application will not pre-process the input data as a separate step and it will be embedded into the model itself. You can find more details in When to Specify Mean and Scale Values.

Regards,

Aznie

[timosy]

Its nice explanation! thanks 

[IntelSupport]

Hi hep77to,

This thread will no longer be monitored since this issue has been resolved. If you need any additional information from Intel, please submit a new question. 

Regards,

Aznie