Use Deep Learning to Assess Palm Tree Health - 2 Small Questions

[RileyBreen]

Hi,

Have gone through the tutorial twice now, and am encountering an error where Geoprocessing Detect Object w Deep Learning tool shows "running" after completing process. The tool has finished running, the data is in it's proper folder, but the status bar still shows running. I have ran it about 6 times now with this .emd and it has done this the last two, but may not be a huge problem, could just be a weird quirk.

Alright, second issue: have many people completed this tutorial? I have ran detect object tool with three different variations: 500 or so samples, 25 epochs ; 800 or so samples, 50 epochs ; and the .emd supplied with the tutorial.

Still have not achieved results comparable to tutorial "finished product" screenshot. Though my model has been okay in deciding which areas do and do not contain palm trees, it seems sporadic, with multiple trees detected where there is one, or a series of trees in a row with only 1 or 2 detected, and the odd ghost palm. Have tried several sets of parameters, as said with the 3 different .emd files. 

Would appreciate any feedback or suggestions,

Thanks

[Wimalavan_Iersel]

Hi Riley,

I have just finished the tutorial as well. I have similar issues with the results, it looks very random. I got 569 training objects, so a bit less, but it should be enough according to the tutrial.

My result looks like this (see attachment).

Would be great if someone could give us some suggestions why our results do not even come close to the screenshot in the tutorial: Use Deep Learning to Assess Palm Tree Health | Learn ArcGIS .Section Palm Tree Detection nr 7

Cheers,

Wimala

[FlorenciaPineiro]

Hi Wimala,

I have the same issue you have reported. Someone may help us?

Regards,

Florencia 

[Maisa]

Hi Riley,

Have completed the tutorial few times. Gave more than 800-1000 object samples with given 25 epochs and also used .emd given with the tutorial. And I had results with multiple trees detection where there is only one palm, and with only 1 or 2 detection  where there were series of trees (as yours). 

Then I decided to try using the other .tif I had and unfortunately my running process was failed all the time so I was not able to even get my .emd file.

Me also would be grateful if someone could give us some suggestions why our results do not even come close to the screenshot in the tutorial.

Kind regards,

Maisa.

[jshiff9898]

I understand that this comment may be redundant, but I too have had similar issues. I had been working on another deep learning project in Arc Pro and was using this tutorial simply to confirm that this workflow is possible. I used about 500 samples, which may not have been enough, but I used the suggested bookmarks so I figured this would at least work a little bit. However, my training returned with an average precision of 0.2 and I was not able to detect a single palm tree. It would be much appreciated if someone who has successfully used this workflow could give us a couple of potential troubleshooting options because it is difficult to tell where I went wrong despite following this tutorial word for word.