Working with deep learning data exported for Mask RCNN

[CalvinEchols1]

Hello everyone!

I am working on training an mask rcnn model to detect an object from a feature layer I have. I have aerial imagery and a feature layer with polygons of objects that I want to detect with the mask rcnn. I have run the export data for deep learning geoprocessing tool, and the resulting output is a folder with a few files and two folders. one is called images and contains the image chips. The other is call label and contains a single folder called 1. that folder contains the masks that correspond to the images in the images folder. all of the mask images appear to be solid black for some reason. has anyone else experienced this?

Also I expected the tool to export a 0 folder in the labels for the background behind the mask, but it does not appear to be in the data generated by the geoprocessing tool.

Thanks for any help, and please let me know if I can provide any additional information.

[CaroGru]

Hey ! I have already seen this questions before here : Exporting training data for deep learning tool in ArcGIS Pro? - Geographic Information Systems Stack... 

If you look at the image from the folder called  "1", they are indeed all black. However, if you import them in ArcGis Pro (or any GIS software I guess), you can see that they are not all black, but black and white, the white part beeing the object that you want to train your model to detect.

EDIT : Based on the pseudo, I guess you are the one who answer on GIS Stack exchange. Hopefully, it will help someone else  

[Anonymous User]

Hello,

Adding to what Caro said, the data belonging to the background class does not gets exported.