IMAGINE Discussions

Discuss and share topics of interest using ERDAS IMAGINE the world’s leading geospatial data authoring system.
Showing results for 
Search instead for 
Do you mean 
Reply
Valued Contributor
Posts: 565
Registered: ‎10-27-2015
Accepted Solution

Object Detection using Deep Learning

Dear all,

I am doing some tests with the new Deep Learning Object Detection Algorithm

I have an image (ECW, 3 bands, Worldview-3 pansharpened).

I have created the training dataset  (using ERDAS) using chips and footprint

 

The creation of the training set was ok. I created a new folder with all the chips (in tif format) and the xml (one for each tif).

If I try to use the Initialize Object Detection Operator ( I gave as input the folder with the chips and the xml) but I have this error:

14/12/18 15:38:16 SessionMgr(1828): Running command line: C:/Program Files/Common Files/Hexagon/SpatialModeler/16.5/bin/x64URelease/mlpywrapper_gpu.exe "C:/Program Files/Hexagon/ERDAS IMAGINE 2018/etc/MachineLearning/Operators/initializeobjectdetection.
py" "C:/Users/maldera/AppData/Local/Temp/SPATIAL_MODELER-4ae5-a11f-8566-5693-013892/mlpy_jsoninput_d192c473-02fb-4458-8deb-be729e25d492.Json" "C:/Users/maldera/AppData/Local/Temp/SPATIAL_MODELER-4ae5-a11f-8566-5693-013892/mlpy_jsonoutput_49cc8c7e-484c-4
d2c-a1f8-9ba4fbb101ef.Json"
14/12/18 15:40:05 SessionMgr(1828): Could not understand JSON for type: IMAGINE.File
14/12/18 15:40:05 SessionMgr(1828): Operator execution failed.


Where am I going wrong?

thank you
Regards
Giuseppe M.

Staff
Posts: 2
Registered: ‎10-24-2018

Re: Object Detection using Deep Learning

The error message does not give enough clues on what went wrong. Please zip and send me the training images and associated xml files.

 

best regards, Sam

Valued Contributor
Posts: 565
Registered: ‎10-27-2015

Re: Object Detection using Deep Learning

Thank you very much for your support Sam,

I am attaching my folder with the chips

Highlighted
Staff
Posts: 134
Registered: ‎06-30-2016

Re: Object Detection using Deep Learning

Hi Giuseppe 

 

I run the training successfully with the data your provided me. Some of the xml files have a different path to the image. make sure the path tp the images specified in the xml file are correct.

 

I have sent you the machine intellect I generated via file catalyst.

Valued Contributor
Posts: 565
Registered: ‎10-27-2015

Re: Object Detection using Deep Learning

[ Edited ]

Hello Sam,

thank you for your reply. I have created the chips in one shot.

 

I have copied all the paths in a text  file and it seems to me that they are ok:

<path>c:/share/deep_learning/object_detection_madrid/training_chip/Airplane1.tif</path>
<path>c:/share/deep_learning/object_detection_madrid/training_chip/Airplane2.tif</path>
<path>c:/share/deep_learning/object_detection_madrid/training_chip/Airplane3.tif</path>
<path>c:/share/deep_learning/object_detection_madrid/training_chip/Airplane4.tif</path>
<path>c:/share/deep_learning/object_detection_madrid/training_chip/Airplane5.tif</path>
<path>c:/share/deep_learning/object_detection_madrid/training_chip/Airplane6.tif</path>
<path>c:/share/deep_learning/object_detection_madrid/training_chip/Airplane7.tif</path>
<path>c:/share/deep_learning/object_detection_madrid/training_chip/Airplane8.tif</path>
<path>c:/share/deep_learning/object_detection_madrid/training_chip/Airplane9.tif</path>
<path>c:/share/deep_learning/object_detection_madrid/training_chip/Airplane11.tif</path>
<path>c:/share/deep_learning/object_detection_madrid/training_chip/Airplane12.tif</path>
<path>c:/share/deep_learning/object_detection_madrid/training_chip/Airplane13.tif</path>
<path>c:/share/deep_learning/object_detection_madrid/training_chip/Airplane14.tif</path>
<path>c:/share/deep_learning/object_detection_madrid/training_chip/Airplane15.tif</path>

 

What is the problem that you found?

 

thank you
Regards

Valued Contributor
Posts: 565
Registered: ‎10-27-2015

Re: Object Detection using Deep Learning

Any news about this? Should I open a support ticket?

Thank you very much

Regards

Staff
Posts: 134
Registered: ‎06-30-2016

Re: Object Detection using Deep Learning

Hi Giuseppe 

 

The only thing I did is change the path in the xml file to where I saved the images you sent me. And I run the initialization after that. Re-run it again and you can log a bug if you still reproduce the issue

 

Regards, Sam

Valued Contributor
Posts: 565
Registered: ‎10-27-2015

Re: Object Detection using Deep Learning

Thank you Sam, I have just opened a ticket: 00037776

Valued Contributor
Posts: 565
Registered: ‎10-27-2015

Re: Object Detection using Deep Learning

I solved the problem using a workstation with an NVIDIA graphic card with more memory. 

The computer that I was using and on which the operator does not work has a 2GB nvidia quadro graphic card.

Now I am working on a computer with a 6 GB GEFORCE graphic card and the operator Initialize Object Detection is working fine and it's very fast.
 
Now I am using the operator "Detects Object Using Deep Learning". I am working on the omputer with  6 GB graphic card. It seems to me that the operator it's not using the GPU. Is it a normal behaviour (not using the GPU)?

Thank you

Session Log


22/01/19 11:51:05 SessionMgr(8932): Executing spatial model
22/01/19 11:51:45 SessionMgr(8932): Running command line: C:/Program Files/Common Files/Hexagon/SpatialModeler/16.5/bin/x64URelease/mlpywrapper_gpu.exe "C:/Program Files/Hexagon/ERDAS IMAGINE 2018/etc/MachineLearning/Operators/objectdetectionclassify.py
" "C:/Users/utente/AppData/Local/Temp/SPATIAL_MODELER-44ec-9fd2-bba4-8b8b-011628/mlpy_jsoninput_3cab043e-1eb4-488f-9a56-db15ab66fcd8.Json" "C:/Users/utente/AppData/Local/Temp/SPATIAL_MODELER-44ec-9fd2-bba4-8b8b-011628/mlpy_jsonoutput_a759ce0d-81c0-4f4d-
ac12-85381a7565bf.Json"

 

nvidia_.PNG

Valued Contributor
Posts: 565
Registered: ‎10-27-2015

Re: Object Detection using Deep Learning

Dear all,

I am doing many tests these days and I am experiencing a very strange behaviour of the Initialize Object Detection Operator.

Using the same input (100 tif images + 100xml footprints) and the same computer sometimes the model fails and sometimes not (I am tryng to create a demo workflow).

Have you experienced something similar?

Thank you

Regards