Deep Learning to Assess Palm Tree Health: .emd file error

3314
12
02-03-2020 06:38 AM
deleted-user-stdf5pSqu_qT
New Contributor

I'm following the lesson "Use Deep Learning to Assess Palm Tree Health". I've had some issues trying to train the model and decided to skip ahead to reviewing the model. I'm stuck on the step "Palm tree detection." In this step I'm trying to run the Detect Objects Using Deep Learning tool, using the .emd file provided in the lesson. I get the following error:

I'm using ArcGIS Pro 2.4.3.

 

Does anyone have the same issue? Any suggestions would be appreciated. Thanks!

0 Kudos
12 Replies
KarstenKelm
New Contributor II

In addition, i upgraded to ArcPro 2.5 and now the Python Package Manager shows the "palm-detection" env with an error showing "Version Mismatch" -  I upgraded the Pro version in the env as well, but that did not fix the problem.

0 Kudos
by Anonymous User
Not applicable

That error is likely because the software cannot find the deep learning Python package. We are working on making sure these error messages are more accurate, but in the meantime, make sure you have followed the instructions here:

Install deep learning frameworks for ArcGIS—ArcGIS Pro | Documentation 

Make sure to create a new Python environment by cloning the default Python environment in the ArcGIS Pro settings. 

KarstenKelm
New Contributor II

Thanks for the fast reply!!  The tensorflow-gpu and keras-gpu packages need to be installed, even though I won't be able to utilize a GPU?

0 Kudos
by Anonymous User
Not applicable

@Karsten yep! The tools check to make sure everything has been installed correctly, and the list provided in the doc are just the minimum packages required to run deep learning tools

0 Kudos
KarstenKelm
New Contributor II

yea, I did all that... still getting this:

I don't have a GPU available, could that be part of the problem?

0 Kudos
by Anonymous User
Not applicable
We're wondering if there is a version mis-match here. If ArcGIS Pro 2.5 is installed, you should have the arcgis 1.7.0 package installed (go to active python env in Python Command Prompt and type in "conda list" to checked installed Python packages). Did you try setting up a brand new cloned environment and following the steps in the documentation? You could be using an old version of the arcgis Python library from the previously-installed version of Pro.
0 Kudos
KarstenKelm
New Contributor II

Yes, I started from scratch.  New env and installed all the packages mentioned in earlier posts.

I am running arcgispro 2.5 and arcgis 1.7.0 along with the packages required and described in the tutorial.

( conda install -c fastai -c pytorch fastai=1.0.54 pytorch=1.1.0 torchvision ) as well as scikit-image 0.15.0.  To no avail...

Even though I managed to train the model myself ( with solid help from our ML engineers ), in order to just get through the tutorial, I am trying to load the emd file that came with the downloaded tutorial data ( C:\DeepLearning\Data ), but I am still getting the above mentioned image.

Here is what I am doing:

- start ArcPro 2.5 and the CoconutHealth project

- open the "Detect Objects Using Deep Learing" tool

- Input Raster "Kolovai Palms"

- Output Detected Objects "PalmTrees"

- Model Definition "CoconutTrees.emd" (from C:\DeepLearning\Data)

and this error pops up:

I did find out that the machine I am using has a NVIDIA Quadro M4000 GPU. 

0 Kudos
KarstenKelm
New Contributor II

got it to work.  after editing the (1)  _ssd.py and (2) _data.py files.

These are located here:

...\C\Program Files\ArcGIS\Pro\bin\Python\envs\palm-detection2\Lib\site-packages\arcgis\learn\models\_ssd.py

...\C\Program Files\ArcGIS\Pro\bin\Python\envs\palm-detection2\Lib\site-packages\arcgis\learn\_data.py

We added "arcgis.learn" before each ".model" :

Now I can access the next steps such as padding (56???)-  

0 Kudos
by Anonymous User
Not applicable

Okay, this is helpful. We are updating the Learn Lesson, and we're going to look through some of these issues to make sure it's easier to get set up. Thanks so much for posting your fix! 

0 Kudos