OpenVINO Object Detection Python demo bad results

[anaanic]

Hi all,

I am trying to test the Object Detection Python demo.

python3 open_model_zoo/demos/object_detection_demo/python/object_detection_demo.py   -d CPU   -i /home/user/Downloads/car-detection.mp4   -m /home/user/intel/openvino_2021.4.752/deployment_tools/open_model_zoo/models/public/yolo-v3-tf/yolo-v3.xml -at yolo -o output_%03d.jpg

I downloaded the model via model downloader:

./downloader.py --name yolo-v3-tf

and converted it with model optimizer:

python3 ./model_optimizer/mo.py --input_model /home/user/intel/openvino_2021.4.752/deployment_tools/open_model_zoo/models/public/yolo-v3-tf/yolo-v3.pb --input_shape "[1, 416, 416, 3]" --reverse_input_channels

The results are nowhere near like in the video from the link:

https://docs.openvino.ai/latest/omz_demos_object_detection_demo_python.html

I attached two random output frames.

What am I doing wrong?

Thank you.

[Iffa_Intel]

Hi,

I tried the OpenVINO Object Detection Demo with this video sample.

I converted the model using converter.py with the default shape since I didn't provide any input_shape parameter.

From my side it seems to work properly:

This is the full inferencing command that I used:

 

I recommend using the latest version of OpenVINO if it's feasible for you since I noticed you are using version 2021.4.

Cordially,

Iffa

[Iffa_Intel]

Greetings,

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

Cordially,

Iffa