segmentation

Image Segmentation

Based on the TensorFlow tutorial, it is an image segmentation model that runs on Edge TPU.
Only the CPU of Raspberry pi 4 can execute it (integer quantization model).

Models

Use the modified U-Net. See the TensorFlow tutorial for details.
Following the tutorial, the model generated by post training quantization after training Oxford-IIIT Pet Dataset is as follows.
Train a model with this notebook.

• Integer quantization model (model.tflite).
• Edge TPU model (model_edgetpu.tflite)

Note

The license of segmentation_post_training_quant.ipynb follows the original tensorflow/doc license.

Usage

• image_segmentation_tflite_capture_opencv.py:
  

    # move to edge_tpu/segmentation

    # CPU only (Raspberry Pi 4 + PiCamera)
    $ python3 image_segmentation_tflite_capture_opencv.py --model model/model.tflite --thread=2

    # Edge TPU (Raspberry Pi 4 + PiCamera)
    $ python3 image_segmentation_tflite_capture_opencv.py --model model/model_edgetpu.tflite