Model Optimiser TF Parameters

[nikhil__aditya]

Hi there,

Am learning to convert basic Tensorflow Object detection API models to IR format using model optimizer. When coming across the parameters, i have few confusions and i hopefully look forward they'll be rectified here.

1) Why do we have to use --tensorflow_use_custom_operations_config when we're specifying the topology by giving --tensorflow_object_detection_api_pipeline_config. I tried to run it without specifying --tensorflow_use_custom_operations_config but that's not running, i see it's mandatory. Why?

2) --transformations_config:  A subgraph replacement configuration file with transformations description. Can we use this instead of --tensorflow_use_custom_operations_config as they both specify the same file. And what transformations mean here?!

Thank you,

Aditya Nikhil

[SIRIGIRI_V_Intel · Accepted solution]

Hi Aditya,

Since there are different variations in the same topology and versions for example ssd v1.14, ssd v1.15, ssd_v2 etc. The Architecture is one but the nodes are different. So the custom operations for the respective versions are defined in the corresponding json files located in the <path_to_model_optimizer>/extensions/front/tf/

--tensorflow_use_custom_operations_config has been renamed to --transformations_config. The old command line parameter is deprecated and will be removed in the future releases.

It is used to replace the specific operations to the OpenVINO format

Regards,

Ram prasad