Arcgis Pro 2.6 Classify Pixel Using Deep Learning (strange result)

[YuraBilinsky]

Hi, community. 

I've change version of ArcPro to 2.6 recently. So needed to reinstall deep  learning tools to

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

tensorflow-gpu=2.1.0

When i started to classify my image i received that:

 But when i back to old versions of libraries for deep learning 

arcgis=1.8.2

scikit-image=0.15.0

pillow=6.2.2

libtiff=4.0.10

fastai=1.0.54

pytorch=1.1.0

torchvision=0.3.0

tensorflow-gpu=1.14.0

All back to normal (using the same model as before)

Maybe some of You had such a problem or maybe know what can call it.

Thank You. Best regards, Oleh

[DanPatterson]

Yura Bilinsky I have moved this to Imagery and Remote Sensing .  You might have a better chance getting a response there

[YuraBilinsky]

Thank You Dan.

Maybe that helps.

Give some details:

1. I used the same type of imagery, same version of Arcpro,  same deep learning model. Only change was libraries pytorch (tensorflow).

2. Also tried on the different type of imagery (simple RGB, non multispectral). Result the same - on new pytorch and tensorflow - result with artifacts as on first figure.

3. Even tried to train a new model on new libraries. Training comes with no problems. But classification causes the same strange result as mentioned before.