Hi Ian,
It is a good idea to try things out on small portion / extent of the data first before running a long job. Instead of setting extent for the gp tool, a more flexible way is to use on-the-fly raster functions. You can add those onto the image service layer in Pro or ArcMap. Then you can pan and zoom to different locations and check out the results in important hot spots before you run it all on TB of pixels.
The SegmentMeanShift raster function (not the gp tool) only takes 3 band 8 bit input. It was developed originally for RGB images. Add the ExtractBandFunction first to pick the 3 bands you want to use. This can be RGB, but you can also create a false color image using the IR band + 2 other bands. It depends on the task, what you want to do or classify in the end. Choose the 3 bands such that things you want to distinguish later are as different in color as possible. On top of ExtractBandFunction, add SegmentMeanShift function and it should work fine.
If you have input 16 bit not 8 bit, add Stretch function to scale pixel values down to 8 bit.
In general, play with ExtractBands and Stretch and their params until you like the intermediate, 3 band 8 bit false color image. Then apply Segment on that result.
Once you are done and happy with your function chain applied to image service, remove the segment raster function and run the segment gp tool on that layer. Alternatively, you can also run the function chain (incl the segment raster function) in the GenerateRasterTool for parallel processing.
Hope that helps. Best, Tom.
