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.

runtime error in model usage

pkhan10
New Contributor I
752 Views

 

 

 

hello 

i have trained ssd object detector with backbone of mobilenetv2

I was able to convert model using model optimizer, using following script

python mo_tf.py --input_model /media/prateek/prateek_space/model_files/person_detection/2020_07_03_mobilenet_v2/inference_graph/frozen_inference_graph.pb --tensorflow_object_detection_api_pipeline_config ~/prateek_space/model_files/person_detection/2020_07_03_mobilenet_v2/inference_graph/pipeline.config --input_shape [1,3,300,300] -o ~/prateek_space/model_files/openvino_model/2020_07_03_person_detection/ --transformations_config extensions/front/tf/ssd_support_api_v1.15.json --reverse_input_channels

 

I am getting following runtime error while using the model, please help

Screenshot from 2020-07-03 13-29-00.png

0 Kudos
1 Solution
Max_L_Intel
Moderator
689 Views

Hi @pkhan10 

Okay, in that case please modify your command in the following way

python mo_tf.py --input_model /media/prateek/prateek_space/model_files/person_detection/2020_07_03_mobilenet_v2/inference_graph/frozen_inference_graph.pb --tensorflow_object_detection_api_pipeline_config ~/prateek_space/model_files/person_detection/2020_07_03_mobilenet_v2/inference_graph/pipeline.config --input_shape [1,300,300,3] -o ~/prateek_space/model_files/openvino_model/2020_07_03_person_detection/ --transformations_config extensions/front/tf/ssd_support_api_v1.15.json --reverse_input_channels

I am able to successfully convert your model and then do a test inference on it with the benchmark app. Please test this with your application.

Best regards, Max.

View solution in original post

0 Kudos
4 Replies
Max_L_Intel
Moderator
733 Views

Hi @pkhan10 

Weights dimensions are {16,3,3,3}, but input is {1,300,3,300}.

Either weights should have the dimensions {16,300,3,3} or input should be {1,3,3,300} (--input_shape [1,3,300,3] in your command since you use NHWC format).

Hope this helps.

Best regards, Max.

0 Kudos
pkhan10
New Contributor I
725 Views

Hey max, 
thanks for responding back
attaching all model files here including pipeline.config
as you can see input shape is (300,300), 
then as per that --input_shape in model optimizer should be [1,3,300,300]
i have attached model files aswell,
Can you please tell me write approach 

0 Kudos
pkhan10
New Contributor I
723 Views

model weights and pipeline

0 Kudos
Max_L_Intel
Moderator
690 Views

Hi @pkhan10 

Okay, in that case please modify your command in the following way

python mo_tf.py --input_model /media/prateek/prateek_space/model_files/person_detection/2020_07_03_mobilenet_v2/inference_graph/frozen_inference_graph.pb --tensorflow_object_detection_api_pipeline_config ~/prateek_space/model_files/person_detection/2020_07_03_mobilenet_v2/inference_graph/pipeline.config --input_shape [1,300,300,3] -o ~/prateek_space/model_files/openvino_model/2020_07_03_person_detection/ --transformations_config extensions/front/tf/ssd_support_api_v1.15.json --reverse_input_channels

I am able to successfully convert your model and then do a test inference on it with the benchmark app. Please test this with your application.

Best regards, Max.

0 Kudos
Reply