Using Yolo Tensorflow for inteference openvino r2

[abhi1]

I converted my darknet yolo to tensorflow & then xml openvino .....tiny model following this guide: https://github.com/PINTO0309/OpenVINO-YoloV3

for creation of .xml I have used open r2. Now I want to do inference with TensorFlow YOLO...xml file

to do detection on Raspberry PI using webcam. Any instructions?

[JesusE_Intel · Accepted solution]

Hi abhi,

The OpenVINO toolkit for Raspberry Pi does not include the Object Detection Sample for YOLO. However, you should be able to copy the python sample from the OpenVINO toolkit for Linux and run it on your Raspberry Pi.

I am attaching the sample code included in the OpenVINO toolkit for Linux R2 2019. You can run the sample code on the Raspberry Pi with a camera with the following command.

python3 object_detection_demo_yolov3_async.py -m <your model.xml> -i cam -d MYRIAD

Hope this helps!

Regards,

Jesus

[JesusE_Intel]

Attaching the sample code from the OpenVINO toolkit for Linux R2 2019.

[abhi1]

Can you also provide me object_detection_demo_yolov3 tiny .........tiny ?

[JesusE_Intel]

The sample demo I provided should be able to run both YOLOv3 and tiny YOLOv3. Give it a try and let me know the outcome.

Regards,

Jesus

[abhi1]

sure. I'll let you know

[abhi1]

something wrong with predictions. Do you want me to share you the model & required stuff?

[JesusE_Intel]

Hi abhi,

What commands did you use to convert your model to IR format? Did you follow the Steps listed on the OpenVINO documentation for converting the weights to Tensorflow?

You command should look similar to this:

python3 convert_weights_pb.py --class_names <labes/names file> --data_format NHWC --weights_file <weight file> --tiny

To convert the TensorFlow model (pb) to IR, use the following command:

python3 <OpenVINO>/deployment_tools/model_optimizer/mo_tf.py --input_model <tensorflow model> --tensorflow_use_custom_operations_config <json file> --batch 1 --data_type FP16 --reverse_input_channels

If you converted the model differently, try to convert it again with the commands suggested above. Please send me your .pb file and .json file if you continue to have issues.

Regards,

Jesus