Modify parameters for SSD object detection trained with tensorflow

[nnain1]

For SSD training in Tensorflow, I have modified aspect ratios and feature layers. Will they be reflected in OpenVino conversion? 

I am not taking out from 6 layers (default SSD has 6 layers for feature extraction).

Aspect ratio is in config file. Is it taken care in conversion?

How about feature layers? How can I make sure it is reflected in conversion?

[Munesh_Intel]

Hi Nyan Naing,

Model Optimizer supports two types of image resizer, with one of them being keep_aspect_ratio_resizer.

More information is available at the following links:

https://docs.openvinotoolkit.org/2020.3/_docs_MO_DG_prepare_model_convert_model_tf_specific_Convert_Object_Detection_API_Models.html#tf_od_custom_input_shape

https://docs.openvinotoolkit.org/2020.3/_docs_MO_DG_prepare_model_convert_model_tf_specific_Convert_Object_Detection_API_Models.html#keep_aspect_ratio_resizer_replacement

https://docs.openvinotoolkit.org/2020.3/_docs_MO_DG_prepare_model_convert_model_tf_specific_Convert_Object_Detection_API_Models.html#important_notes_about_feeding_input_images_to_the_samples

Additionally, detailed explanations of model conversion process for SSD topologies are available at the following link:

https://docs.openvinotoolkit.org/2020.3/_docs_MO_DG_prepare_model_convert_model_tf_specific_Convert_Object_Detection_API_Models.html#ssd_single_shot_multibox_detector_topologies

Regards,

Munesh