Here is my issue : I have trained a TF2 model on the InceptionResNetV2 topology and converted it to the usual .xml and .bin file with the Model Optimizer included in OpenVino. So far no issue
"[ SUCCESS ] Generated IR version 10 model.
[ SUCCESS ] XML file: /home/wm/docker/Cortex2/age_auto/output_model/model_age.xml
[ SUCCESS ] BIN file: /home/wm/docker/Cortex2/age_auto/output_model/model_age.bin
[ SUCCESS ] Total execution time: 67.71 seconds.
[ SUCCESS ] Memory consumed: 1698 MB. "
When I try to load this model I have the following error : "ERR MOVIDIUS EXCEPTION The support of IR v6 has been removed from the product. Please, convert the original model using the Model Optimizer which comes with this version of the OpenVino to generate supported IR version."
Issue is : I have converted the model and try to execute it from a container made with this same docker image: openvino/ubuntu18_dev with the tag latest... I trust he should agree with himself so I don't understand my mistake.
Could someone give a hand please ?
May I know what data type that you had used when converting the model? did you specify FP16?
By default (without any definition) the model would be converted into FP32.
If you are trying to use that model with NCS2 you will need the converted model in FP16 format.
It didn't change anything. I optimized the tensorflow's .pb with this command :
"python3 deployment_tools/model_optimizer/mo_tf.py --input_model /home/wm/docker/Cortex2/age_auto/output_model/model_age.pb --output_dir /home/wm/docker/Cortex2/age_auto/output_model/ --b 5 --data_type FP16 -s 255 --reverse_input_channels"
When I try to load the models here is the error :
"ERR MOVIDIUS EXCEPTION The support of IR v6 has been removed from the product. Please, convert the original model using the Model Optimizer which comes with this version of the OpenVINO to generate supported IR version."
Which OpenVINO version you are currently using?
It is recommended for you to use the latest version 2021.4, Please help to convert using the model optimizer from this version.
If you are using an outdated version, the biggest chance is there are something that already been deprecated.
Plus, for affirmation, ensure that you are using NCS2 and not the older compute stick: