Making predicted building footprints as accurate as possible

[Murtaza_Mir]

Hi,

I am using Deep Learning to predict/identify building footprints. I also use regularize building footprints tool to polish the boundaries of the footprints. But as I see, it is still not accurate enough. Is there a way to further refine the footprints and make them as accurate as possible?

I am using ArcGIS Pro 3.2 version.

Thank you 

[BobBooth1]

Have a look at this blog post:

https://www.esri.com/arcgis-blog/products/arcgis-pro/analytics/multiresolution-object-detection-with...

[Murtaza_Mir]

Hi Bob,

Thank you for your response. I tried running the Multi Resolution Deep Learning tool and I am getting this Python error:

Traceback (most recent call last):
  File "<string>", line 465, in execute
AttributeError: 'NoneType' object has no attribute 'split'

 Did you experience this error as well?

Thank you

[R_Alouta]

Hi Murtaza,

Can you please provide more information? What parameters are you using with the tool? Is this the only message you are getting? What type of imagery are you using? Any information can be helpful here.

[Murtaza_Mir]

Hi,

I was able to fix this error by specifying the extent. Apparently, if we don't specify the extent in the Environment tab of the tool, then this error gets thrown up.

However, I used this tool on a small imagery data. I believe it won't scale up well if I use more data. Apart from that, even this tool is not precise in its predictions. 

Can you please tell me if there is a way to make predictions as accurate as possible? Any other post processing steps maybe? I make the predictions and I also use Regularize Building Footprints tool. But then again, the boundaries are not precise and there is always some error. I was wondering what other tools/techniques can be used to make the building footprints near 100%