Saving a Object Detection API SSD in TF.SavedModel format

PaulS · January 4, 2022, 2:48am

Hi All,

The “TF Object Detection API” has this colab:

tensorflow/models/blob/master/research/object_detection/colab_tutorials/eager_few_shot_od_training_tf2_colab.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "colab_type": "text",
        "id": "rOvvWAVTkMR7"
      },
      "source": [
        "# Eager Few Shot Object Detection Colab\n",
        "\n",
        "Welcome to the Eager Few Shot Object Detection Colab --- in this colab we demonstrate fine tuning of a (TF2 friendly) RetinaNet architecture on very few examples of a novel class after initializing from a pre-trained COCO checkpoint.\n",
        "Training runs in eager mode.\n",
        "\n",
        "Estimated time to run through this colab (with GPU): \u003c 5 minutes."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {

This file has been truncated. show original

That shows how to build & train a ssd_resnet50_v1_fpn_640x640_coco17_tpu-8 model using ducky images. The colab runs great, but after training, I wish to know how to save this model in the common TF.SavedModel format (Using the SavedModel format | TensorFlow Core).

I have tried using all versions of the ‘exporter_xxx.py’ scripts, and everything gives the same type of errors:
WARNING:absl:Found untraced functions such as
ssd_meta_arch_layer_call_fn,
ssd_meta_arch_layer_call_and_return_conditional_losses,
WeightSharedConvolutionalBoxPredictor_layer_call_fn, WeightSharedConvolutionalBoxPredictor_layer_call_and_return_conditional_losses,
ssd_meta_arch_layer_call_fn while saving (showing 5 of 525). These functions will not be directly callable after loading.

It does produce a saved model, but the model is not usable, giving the error:
“Found zero restored functions for caller function”
When you try to use it.

Does anyone have any code that can be added on to the end of the eager_few_shot_od_training_tf2_colab to save the trained model in the common TF saved model format?

Any help would be appreciated.

Thanks.

Laxma_Reddy_Patlolla · December 11, 2023, 10:41pm

Hi @PaulS ,

Given the recent updates in the TensorFlow Model Garden codebase, it’s advisable to test the latest version to check if the mentioned issue has been resolved. If you are still encountering the problem, consider using the most recent codebase.
Thanks