Train Deep Learning Model Not recognizing Training folders

[NathanJennings]

I am on Pro 2.5.0

I followed the install instructions as per the Pro Help:  Install deep learning frameworks for ArcGIS—ArcGIS Pro | Documentation 

I am using the sample image for swimming pools found here:  https://www.arcgis.com/apps/MinimalGallery/index.html?appid=99c39f7512d54881bc365583c76c7da6  for the PyTorch swimming pool example.

I created about 10 samples using the Label Objects for Deep Learning.

I saved the samples and the labels and then ran the Export Training data which created the images and labels folder, the other files required for the next step to train the model.

The input folder is never recognized (which is:  \deep_learning\redlands_img that contains the above folders and files).  I get the following error and cannot seem to fix it.  Any help is appreciated.  I am including my zip file of output.

Also, the documentation for the specific tools and the overview don't really provide enough details to set tool parameters or suggestions for different types of scenarios.  Some updated material would be most helpful for the deep learning community that support Pro 2.5.

Thank you,

Nate

[MasitaManessa]

i have the same problems. Have you solve this problems. 

thanks

[KyleC]

Hi Masita, thank you for reaching out on GeoNet. This issue was logged as a bug and the current bug status is 'Implemented', which means the defect has been addressed for the next software release. As a workaround for now, for the Export Training Data For Deep Learning tool, under Meta Data Format, choose any format other than KITTI Labels, such as PASCAL Visual Object Classes. Please refer to this page for more information. 

[PriyankaTuteja]

Hi,

Could you please try creating a new deep learning environment using the steps below:

First, make sure that you have your default environment arcgispro-py3 set active. If not, open Python command prompt and run the command: proswap arcgispro-py3 

Note: You might have to close and reopen the Python command prompt to see the correct path. 

Then create a new environment using this command: conda create --name deeplearning --clone arcgispro-py3

To activate this environment, run: activate deeplearning

Further to install deep learning libraries, run the below command:

conda install -c esri -c fastai -c pytorch arcgis=1.8.2 scikit-image=0.15.0 pillow=6.2.2 libtiff=4.0.10 fastai=1.0.60 pytorch=1.4.0 torchvision=0.5.0 --no-pin

Now run: proswap deeplearning. You can now try running Train deep learning tool using Pro. Alternatively, you can also launch a notebook server using the command: Jupyter notebook

you can follow the training steps given in part 2 of

Detecting Swimming Pools using Satellite Imagery and Deep Learning | ArcGIS for Developers. 

Please let me know if you face issues.

Regards,

Priyanka

[ToddBrown]

I was having a similar issue in a different Esri training module:  Use deep learning to assess palm tree health.  I created a new deep learning environment using your steps and the problem was solved.  Thank you.p learning to assess palm tree health

[RDAltarez]

I wonder @PriyankaTuteja if you have an update for this? I am using ArcGIS pro 3.0. I tried to follow the steps you have indicated in the above to no avail.

[PriyankaTuteja]

@RDAltarez Please try using the following commands  to create deep learning environment in Pro 3.0

conda create --clone arcgispro-py3 --name dlenv

proswap dlenv

conda install -c esri deep-learning-essentials

[RDAltarez]

Hi @PriyankaTuteja, thank you for your reply. I have solved my problem. There's a new Jupyter installed together with the ArcGIS pro update. So i just opened it and it works fine. Thanks again for this reply.