models icon indicating copy to clipboard operation
models copied to clipboard

Published 20 hours ago • verified publisher icon tensorflow

Implementation of ssd_mobilenet_v2_quantized (where is the quantizer?)

Open Petros626 opened this issue 2 years ago • 0 comments

Hello,

I am looking for the code or files that quantizes the weights for the architecture ssd mobilenet v2_quantized config file (Original architecture (Python code) from which the config files were created).

I have found the following files:

  • https://github.com/tensorflow/models/blob/627d65d5fa13713515cdcb1a4ea4ff28085adb21/research/object_detection/models/keras_models/mobilenet_v2.py#L16
  • https://github.com/tensorflow/models/blob/627d65d5fa13713515cdcb1a4ea4ff28085adb21/research/object_detection/models/ssd_mobilenet_v2_feature_extractor.py#L16

I have found the model architecture and the feature extractor for this, however it is not clear to me how the model is retrained quantized when loading the quantized config file during training.

Thank you very much for any information

Petros626 avatar Feb 13 '23 12:02 Petros626