Classify Pixels Using Deep Learning Produces Empty Output

[dhan]

Hello,

I'm currently attempting to follow this model guide here: Use the model—ArcGIS pretrained models | Documentation

The guide uses this model here: Land Cover Classification (Aerial Imagery) - Overview

This is the tif file that I originally started with:

However, since this tif file has 4 bands and was in state plane feet, I projected (Project Raster) the layer so that it was in meters and then extracted bands (Extract Bands) 1,2, and 3. That result is shown here:

This layer was then used as the input for the Classify Pixels Using Deep Learning tool. The parameters were set as the guide had shown them. This is how I set the tool up:

However, my output from that tool returns a raster layer that only has 0 or 255 values and not the distinct land cover classes. Here is what the output layer properties look like:

And this is the output layer visualized by stretch:

Is there some step I am missing? Or maybe there is some sort of mismatch between my input and what the model expects? 

Thanks in advance.

[PavanYadav]

@dhan

hi @dhan Can you please share a small portion of your input raster? Also, just for a test, could you try an object detection model? Depending on the objects in your image, you might try Text SAM or Building Footprints detections. I'd like to see if this is a software or a model problem. Thanks!

[dhan]

Hi @PavanYadav ,

Thanks. The resolution is very fine so it exceeds the maximum file size limit. Is there an alternate way of sharing the sample data? Or I can try to extract a very small extent to be under 9MB, but I am not sure if the model would work then.

[PavanYadav]

@dhanCould you please email me at pyadav AT esri DOT com with a link to the data on OneDrive or another password-protected drive? Or, you can just email and we can set up meeting if you're in PST zone.