Output shape mismatch in object detection using yolov3 openvino

[Ajsal]

I'm trying openvino for object detection using yolov3. I have downloaded the model using the script located in /opt/intel/openvino_2021/deployment_tools/tools/model_downloader. Converted the downloaded model to intermediate representation using the script /opt/intel/openvino_2021/deployment_tools/model_optimizer. The command used is as follows,

python3 mo_tf.py -b 1 --input_model /opt/intel/openvino_2021/deployment_tools/tools/model_downloader/public/yolo-v3-tf/yolo-v3.pb --tensorflow_use_custom_operations_config /opt/intel/openvino_2021/deployment_tools/tools/model_downloader/public/yolo-v3-tf/yolo-v3.json --data_type FP16 --output_dir /home/ubuntu/openvino_models/ir/public/yolo_v

But object detection script (/opt/intel/openvino_2021.4.752/deployment_tools/inference_engine/samples/cpp/object_detection_sample_ssd/main.cpp) throws error as given below,

[ INFO ] InferenceEngine:
IE version ......... 2021.4.2
Build ........... 2021.4.2-3974-e2a469a3450-releases/2021/4
Parsing input parameters
[ INFO ] Files were added: 1
[ INFO ] /home/ubuntu/Desktop/Models/tensorflow-yolo-v3/2.jpg
[ INFO ] Loading Inference Engine
[ INFO ] Device info:
CPU
MKLDNNPlugin version ......... 2021.4.2
Build ........... 2021.4.2-3974-e2a469a3450-releases/2021/4

[ INFO ] Loading network files:
[ INFO ] /home/ubuntu/openvino_models/ir/public/yolo_v3/yolo-v3.xml
[ INFO ] Preparing input blobs
[ INFO ] Batch size is 1
[ INFO ] Preparing output blobs
[ ERROR ] Output item should have 7 as a last dimension
[1] + Done "/usr/bin/gdb" --interpreter=mi --tty=${DbgTerm} 0<"/tmp/Microsoft-MIEngine-In-hpcpwfz2.o5d" 1>"/tmp/Microsoft-MIEngine-Out-xxzadbkj.g0x"

Can somebody help to resolve this issues?. I have been trying my head around this for hours. Any help would be greatly appreciated. Thanks.

[Zulkifli_Intel]

Hello Ajsal,

Thank you for reaching out to us.

The Object Detection C++ Sample SSD does not support the YOLO architecture model. You can run the sample with the SSD models.

If you want to use the yolo-v3-tf model, we recommend you use the Object Detection Python Demo that supports the YOLO architecture model.

Sincerely,

Zulkifli

[Vladimir_Dudnik]

@Ajsal  Open Model Zoo also provide C++ Object Detection Demo and Multi-Channel Object Detection Yolov3 Demo, which support yolo-v3 as well, just in case you are looking for C++ implementation in particular.

[Zulkifli_Intel]

Hello Ajsal,

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.

Sincerely,

Zulkifli