Intel® Distribution of OpenVINO™ Toolkit
Community assistance about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all aspects of computer vision-related on Intel® platforms.
6446 Discussions

Openvino DL workbench visualize output not working

Hasanul_K_Intel
Employee
1,608 Views

Hi,

I have trained a YOLOV4 model on a custom dataset and have converted it to tensorflow .savedmodels format following the instructions from https://docs.openvino.ai/latest/workbench_docs_Workbench_DG_Tutorial_Import_YOLO.html

 

I was able to import this model and quantize it to FP16 in DL WB. However, when I try to visualize the output of the model by selecting an image, it doesn't work. It just keeps loading forever. Below are the screenshots of my setup. Now, I have tried both object detection and classification menu for this and neither of them worked. YOLO is mentioned under object detection, but it does perform classification as well and I actually would love to see the importance map of the pixels to understand the few cases where the model is not classifying it right. 

 

Any suggestion is highly appreciated. Please let me know if I need to give more details.

 

Hasanul_K_Intel_0-1642287912229.pngHasanul_K_Intel_1-1642287935847.png

The blue circle on the right has been loading for a day now and nothing happens. 

 

Best regards,

Hasanul

0 Kudos
9 Replies
Alexander_Suv_Intel
1,580 Views

Hi Hasanul,

 

Thank you for reaching out. In order to investigate an issue we need additional details. Could you please provide a model in darknet (.weights and .cfg) and tensorflow (.savedmodel) formats?

YOLO models are supported as object detection only in DL Workbench. So unfortunately importance map visualization is not available for YOLO models at the moment.

 

Best regards,

Alexander

0 Kudos
Hasanul_K_Intel
Employee
1,553 Views

Hi Alexander,

 

Thanks for the explanation on the map visualization. I can attach the .cfg file but cannot attach the other files as they are beyond the file size limit of 71 MB. 

Please let me know how could I share those files here and I will share all of them together.

 

Best regards,

Hasanul 

0 Kudos
Alexander_Suv_Intel
1,521 Views

Hi Hasanul,

 

Could you please use any convenient for you share option.

 

Best regards,

Alexander

0 Kudos
Hasanul_K_Intel
Employee
1,508 Views

Hi Alexander,

 

I have shared in onedrive the files that you need. Please use the link below to get the files. Please let me know if you need any other files.

 

https://intel-my.sharepoint.com/:f:/p/hasanul_karim/EsGXagYHED1Mrnz_dOFzsJkB8Mmd7jg5ZjXY9Lvevg_duQ?e=h6gqXN

 

Best regards,

Hasanul

0 Kudos
Alexander_Suv_Intel
1,501 Views

Hi Hasanul,

 

Thanks for providing your model. Just checked files and have several questions.

1. The savedmodel directory size is less than 8mb. Is it correct? Just double checking because you mentioned that your model is quite large and doesn't fit to 70mb limit.

2. Could you please also provide .weights file in darknet format?

0 Kudos
Alexander_Suv_Intel
1,495 Views

Thanks for sharing .weights file. It's seems we have everything we need. Please give us some time for investigation. I'll get back to you when I have some results.

 

Best regards,

Alexander

0 Kudos
Hasanul_K_Intel
Employee
1,490 Views

Hi Alexander,

 

Thanks for your prompt response. Just to confirm, I also checked the .savedmodel folder, the files in the directory are small, but there is a folder named variables inside there that is 245 MB. I think everything is there. Please let me know if you find any issue with the files.

 

Looking forward to hearing from you soon.

 

Best regards,

Hasanul

0 Kudos
Alexander_Suv_Intel
1,456 Views

Hi Hasanul,

 

Thank you for sharing .savedmodel, I was able to find 245 MB file in a variables directory.

After investigation we can confirm that the issue is exist in version 2021.4.2 .

0 Kudos
Munesh_Intel
Moderator
1,417 Views

Hi Hasanul,

Our development team is working on this bug, and this issue will be fixed in the next OpenVINO release.

 

This thread will no longer be monitored since we have provided a solution. If you need any additional information from Intel, please submit a new question.

 

 

Regards,

Munesh


0 Kudos
Reply