Properly set image_dim value in MLP classifier pipeline

uncharted-distil / distil-auto-ml

Distil Automated Machine Learning Server

Apache License 2.0

2 stars 1 forks source link

Properly set image_dim value in MLP classifier pipeline #245

Closed cdbethune closed 4 years ago

cdbethune commented 4 years ago

We currently can only run the MLP classifier with big earth data since the image_dim value is not being set based on the underlying remote sensing data. We either have to figure out the proper dimension at the time the pipeline is constructed, or have it be determined at run time, and passed along as metadata.

cdbethune commented 4 years ago

Our interpretation of gradcam results expects a 4x4 matrix back, so the solution to our problem actually ends up being that we want to force the image_dim value to always be 4. No need to include the source image size.