Deep Learning KITTI Labels BUG

1363
3
04-27-2020 12:49 PM
Bahram-Jomehzadeh
New Contributor III

I've been trying Train Deep Learning Model (in order to building footprint extraction from WorldView 2 image) using exported training data with KITTI Labels meta data format but every time add the output Image Chips folder faced this error.

ERROR 032659 updateParameters Execution Error: Traceback (most recent call last):
 File "c:\program files\arcgis\pro\Resources\ArcToolbox\toolboxes\Image Analyst Tools.tbx#TrainDeepLearningModel_ia.UpdateParameters.py", line 4, in <module>
 File "c:\program files\arcgis\pro\Resources\ArcToolbox\toolboxes\Image Analyst Tools.tbx#TrainDeepLearningModel_ia.UpdateParameters.py", line 62, in updateParameters
IndexError: list index out of range

Any opinion?

GIS Consultant
Proficient in best practices and uses of Esri's ArcGIS technologies.
0 Kudos
3 Replies
VinayViswambharan
Esri Contributor

Hey Bahram - 

  1. Please use RCNN as the metadata format when exporting your training samples, as you are trying to delineate building footprint (and not just derive bounding polygons around buildings).
  2. Also when running the train deep learning model tool, explicitly, change the model type parameter (it will be blank by default which is causing your error messafe)and that should solve your problem

Regards,

Vinay

Bahram-Jomehzadeh
New Contributor III

How are you Vinay.

I actually had tried the rest of the models including MaskRCNN - Object detection and unfortunately the best result I've achieved (applying MaskRCNN - Object detection) that is not nothing satisfying!

I was wondering if using KITTI Labels would be better result!?

GIS Consultant
Proficient in best practices and uses of Esri's ArcGIS technologies.
0 Kudos
CaroGru
New Contributor

From what I can see on the second screenshot you provided, your training data are the outlines of the buildings. For MaskRCNN to work better, I would change your training data to polygons (a polygon = a building).

0 Kudos