I want to use the Export Training Data For Deep Leaning tool to export Classified Tiles to use in the arcgis.learn `prepare_data` function and then the UnetClassifier. As a note, I am running Arcgis Pro on my mac using Parallels.
My raster data set is the World_Imagery layer at https://services.arcgisonline.com/ArcGIS/rest/services/World_Imagery/MapServer.
My vector dataset has 43,071 features, which covers my area of interest, Fort Worth. "Output No Feature Tiles" is set to False.
When I try to run the Export Training Data For Deep Leaning on all 43,071 features, progress stalls at 100% and then eventually throws error 9999. In this case, my "images" and "labels" folder each has around 1140 numbered tiffs. The map.txt file counts up to somewhere in the 1000s.
I have tried running the tool on a subset of 100 features of my vector, labeled data set. In this case, progress stalls at 100% and reminded stalled for around 12 hours. I ended up with 1531 tiffs in my "labels" and "images" folders. My map.txt files counts up from "000000000000.tif" to "000000000604.tif" then it starts over and counts from "000000000000.tif" to "000000001028.tif"
Here are my details for the process run with the 100-feature subset:
Export Training Data For Deep Learning
=====================
Parameters
Input Raster World_Imagery
Output Folder \\Mac\Home\Documents\ArcGIS\Projects\cars\trainingdata\1
Input Feature Class Or Classified Raster Or Table TRN_EDGE_OF_PAVEMENT_subset
Image Format TIFF
Tile Size X 256
Tile Size Y 256
Stride X 128
Stride Y 128
Output No Feature Tiles ONLY_TILES_WITH_FEATURES
Metadata Format Classified_Tiles
Start Index 0
Class Value Field ROAD_GROUP_NUM
Buffer Radius 0
Input Mask Polygons TRN_EDGE_OF_PAVEMENT
Rotation Angle 0
Reference System MAP_SPACE
Processing Mode PROCESS_AS_MOSAICKED_IMAGE
Blacken Around Feature NO_BLACKEN
Crop Mode FIXED_SIZE
Additional Input Raster
Instance Feature Class
Instance Class Value Field
Minimum Polygon Overlap Ratio 0
=====================
Messages
Start Time: Friday, July 21, 2023 9:53:25 PM
Distributing operation across 2 parallel instances.
NumTilesWritten = 15, NextTileIndex = 55
NumTilesWritten = 447, NextTileIndex = 605
NumTilesWritten = 0, NextTileIndex = 0
NumTilesWritten = 196, NextTileIndex = 381
NumTilesWritten = 217, NextTileIndex = 726
NumTilesWritten = 35, NextTileIndex = 855
NumTilesWritten = 10, NextTileIndex = 906
NumTilesWritten = 83, NextTileIndex = 1029
could be the machine, or not enough resources... did you see the recommendations here
Run ArcGIS Pro on a Mac—ArcGIS Pro | Documentation
and running deep learning would require more resources
Do we have any solutions here? I ran into the same thing with ArcGIS Pro 3.3.1.