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.

person-detection-0203 detection issues

pkhan10
New Contributor I
472 Views

Hello,
I have finetuned person-detection-0203 on my dataset using openvino training extensions 

but when i test the model i can see the offset in the model output

check the inline image below
on same model pipeline person-detection-0202 is working perfectly

 

pkhan10_0-1624794161273.png

 

just want to understand how i can correctly use output of person-detection-0203

 dict_keys(['TopK_1181.0', 'TopK_1415.0', 'TopK_1649.0', 'TopK_1883.0', 'TopK_947.0', 'boxes', 'labels'])

as of now i have just used boxes and labels for developing this output

i have assumed boxes are in this format  [x_min, y_min, x_max, y_max, conf]

 

 

0 Kudos
3 Replies
Wan_Intel
Moderator
425 Views

Hi Prateek Khandelwal,

Thank you for reaching out to us.


We are investigating this issue and will update you at the earliest.


Regards,

Wan


0 Kudos
Zalesskaya_Galina
400 Views

Hello Prateek Khandelwal,

You are right, the bbox output format is  [x_min, y_min, x_max, y_max, conf] for person-detection-0202 and person-detection-0203 as well. 

I could not reproduce the error, so could you please provide information, which tool did you use for visualization? If you used object_detection_demo, could you provide a hash commit of the used version? Please make sure, that your version has this fix. 

Also, do results during fine-tuning satisfy you, I mean, are they greater than 0 mAP?

 

0 Kudos
Wan_Intel
Moderator
373 Views

Hi Pkhan10,


This thread will no longer be monitored since we have provided a solution.


For your information, if you are unsatisfied with the model quality, you may prepare datasets, evaluate pre-trained model and run fine-tuning until you get acceptable quality metrics values on your data.


Please refer to here for more information.


If you need any additional information from Intel, please submit a new question.


Regards,

Wan


0 Kudos
Reply