Validation encountered error and export error

[Ramson]

Hi Team,

We have trained a object detection model using updated eIQ portal 2.2.14.When trying to validate the model, we are facing the following error. 

Trace: ShowError -> {"type":"error","message":"Validation encountered error: "}
    at N (C:\nxp\eIQ_Toolkit_v1.1.8\resources\app.asar\background.js:1:105046)
    at IpcMainImpl.<anonymous> (C:\nxp\eIQ_Toolkit_v1.1.8\resources\app.asar\background.js:1:110550)
    at IpcMainImpl.emit (events.js:327:22)
    at Object.<anonymous> (electron/js2c/browser_init.js:161:9692)
    at Object.emit (events.js:315:20)
Error Details -> undefined
Trace: ShowError -> {"type":"error","message":"Validation encountered error: "}
    at N (C:\nxp\eIQ_Toolkit_v1.1.8\resources\app.asar\background.js:1:105046)
    at IpcMainImpl.<anonymous> (C:\nxp\eIQ_Toolkit_v1.1.8\resources\app.asar\background.js:1:110550)
    at IpcMainImpl.emit (events.js:327:22)
    at Object.<anonymous> (electron/js2c/browser_init.js:161:9692)
    at Object.emit (events.js:315:20)
Error Details -> "An error occured while converting model"

And also when tried to export the model, similar kind of error occurs. This happens only when trying to export model as rtm file. but not in other formats. 

Trace: ShowError -> {"type":"error","message":"Failed to convert model:"}
    at N (C:\nxp\eIQ_Toolkit_v1.1.8\resources\app.asar\background.js:1:105046)
    at IpcMainImpl.<anonymous> (C:\nxp\eIQ_Toolkit_v1.1.8\resources\app.asar\background.js:1:110550)
    at IpcMainImpl.emit (events.js:327:22)
    at Object.<anonymous> (electron/js2c/browser_init.js:161:9692)
    at Object.emit (events.js:315:20)
Error Details -> "TypeError: Failed to fetch"

I'm attaching a sample deepview project link: defaultvoc.deepview .You can recreate the issue by just training for 1 epoch and try to validate it and export it.

Does anyone have any intuition why this error is happening?  

[Ramson · Accepted solution]

Hi @david_piskula , the issue was solved once after I placed the project in another folder and ran again. I'm not sure whether it was not able to retrieve the project since the path was too long. But anyway, thanks David for your time and effort. Really appreciate the support.

Thanks and regards,

Ramson Jehu K

[david_piskula]

Hi @Ramson,

I'm glad your issue has been resolved. Thanks for letting me know about the fix. We will need to investigate this issue internally to see what potential problems there are with path length.

Regards,

David

[david_piskula]

Hi @Ramson,

I assume this is caused by some incompatibility issues in the converter. I will report this to the development team and let you know what they respond with.

Could you please either upload the project elsewhere or give me rights to access the site and download it? I don't have access at the moment.

Does the model work, if you export it to other formats than RTM? Did you try running it on-device?

Regards,

David

[Ramson]

Hi David,

Could you please either upload the project elsewhere or give me rights to access the site and download it? I don't have access at the moment.

Ok Please try this link : https://drive.google.com/file/d/1-FNUDH4RaAV5XFVl0a5IBsJdtLkhwz0d/view?usp=sharing 

Does the model work, if you export it to other formats than RTM? Did you try running it on-device?

I'm still working on increasing the accuracy of model, which requires validating the model. so I haven't deployed on the device yet.

Thanks and Regards,

Ramson Jehu K

[david_piskula]

This download link works fine, thanks. What model did you train on the dataset? When I tried using the "Object Detection > Balanced > MCU" and trained it for one epoch, the validation worked fine. Are you using a custom model? Can you share the model plugin file for eIQ Portal as well?

[Ramson]

I tried using the "Object Detection > Performance > MCU". Even when i selected balanced, the same error occured. 

[david_piskula]

Interesting. I'm not able to reproduce the error with your project. I tried it with the performance model settings but validation still works fine for me. I noticed your UI has a small difference to mine:

- you have only 5 options on the slider whereas I have 6. Are you sure you are using the eIQ Portal v2.2.14 (eIQ Toolkit v1.1.8) version of the tool?

Can you describe the exact steps you take from opening the project you shared to getting the error? I just open the project, choose the model, don't change any settings whatsoever, let the model train for one epoch and then validate it and it works.

[Ramson]

yes im using v2.2.14 only. 
Steps :
1. open the project
2. Select model -> Detection - > Performance -> MCU 
3. Set batch size to 64
4. Set Default Augmentation - Oriented
5. Set Scale = small, Activation = relu6, Optimizer = Adam

[david_piskula]

I'm not able to reproduce the issue using the defaultvoc project you shared and following the steps you describe. Validation always works fine for me.

Could you share the full logs, please? (%AppData%\eIQ Portal\logs)

Are you using Windows or Linux?

Could you try completely uninstalling all versions of eIQ Toolkit/Portal from your PC and then installing a completely fresh installation again, please?

[Ramson]

Hi @david_piskula 

 I have installed CUDA version 11.2.2 and cuDNN version 8.1.1 which was mentioned in the Tensorflow guide. 

What version of CUDA and cuDNN version do you use?

Thanks and Regards,

Ramson Jehu K

[Ramson]

Could you share the full logs, please?

yes, I have attached the log file. Please check it out.

Are you using Windows or Linux?

I'm using windows 10

Could you try completely uninstalling all versions of eIQ Toolkit/Portal from your PC and then installing a completely fresh installation again, please?

Yes, I tried uninstalling all the versions of eIQ toolkit and reinstalled. But still the issue persists.