E2E-Object-Detection-in-TFLite

This repository shows how to train a custom detection model with the TFOD API (TF2 and TF1), optimize it with TFLite, and perform inference with the optimized model.

About the notebooks

• Training_a_pets_detector_model_within_minutes_with_TFOD_API.ipynb: Shows how to train a custom object detection model on the Pets dataset (non-eager mode).
• Running_inference_with_a_custom_TFOD_API_model.ipynb: Shows how to export a SavedModel graph from the trained checkpoint files, and run inference.
• Object_Detection_in_TFLite.ipynb: Shows how to quantize the original model, generate a TFLite model, and run inference.
• colab_training/ -
  • Fruits_Detection_Data_Prep.ipynb: Shows how to prepare a object detection dataset in order to train it with the TFOD API in non-eager mode.
  • Training_MobileDet_Custom_Dataset.ipynb: Shows how to fine-tune a MobileDet model to detect localize and detect fruits. It also shows how to optimize the fine-tuned model with TensorFlow Lite and run inference with the optimized model.

Note

Training_a_pets_detector_model_within_minutes_with_TFOD_API.ipynb notebook uses Colab to demonstrate the training workflow but does not actually uses the Colab runtime for training. It uses Cloud TPUs.

Model usage on mobile devices

Sample applications (both Android and iOS) are available here.

TFLite model files

Available here (currently available in dynamic-range only). It was generated from the first notebook (Training_a_pets_detector_model_within_minutes_with_TFOD_API.ipynb).

Acknowledgement

I would like to thank the ML-GDE program for providing with GCP credits without which this project wouldn't have been possible.