Convert Model to Edge TPU TFlite Format for Google Coral
Introduction
The Coral M.2 Accelerator with Dual Edge TPU is an M.2 module that brings two Edge TPU coprocessors to existing systems and products with an available M.2 E-key slot.Tensorflow and Pytorch is the most popular deep learning frameworks. So in order to use the Edge TPU, we need to compile the model to Edge TPU format.
This wiki article will guide you through the process of compiling a model and running it on the Google Coral TPU, enabling you to leverage its capabilities for high-performance machine learning applications.
Prepare Hardware
Raspberry Pi 5 8GB | Raspberry Pi M.2 HAT+ | Coral M.2 Accelerator B+M key |
---|---|---|
![]() | ![]() | ![]() |
Install Hardware
Convert Model
Before you start, make sure you have installed the Google Coral TPU to Pi 5 follow the installation guide.
- For Tensorflow Model
- For Pytorch Model
- For Yolo Model
And all process have been tested on Python 3.11.9.
Install Tensorflow
pip install tensorflow
Check tflite_converter
tflite_convert -h
The result should be like this:
2024-07-23 10:41:03.750087: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2024-07-23 10:41:04.276520: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT
usage: tflite_convert [-h] --output_file OUTPUT_FILE [--saved_model_dir SAVED_MODEL_DIR | --keras_model_file KERAS_MODEL_FILE] [--saved_model_tag_set SAVED_MODEL_TAG_SET]
[--saved_model_signature_key SAVED_MODEL_SIGNATURE_KEY] [--enable_v1_converter] [--experimental_new_converter [EXPERIMENTAL_NEW_CONVERTER]]
[--experimental_new_quantizer [EXPERIMENTAL_NEW_QUANTIZER]]
Command line tool to run TensorFlow Lite Converter.
optional arguments:
-h, --help show this help message and exit
--output_file OUTPUT_FILE
Full filepath of the output file.
--saved_model_dir SAVED_MODEL_DIR
Full path of the directory containing the SavedModel.
--keras_model_file KERAS_MODEL_FILE
Full filepath of HDF5 file containing tf.Keras model.
--saved_model_tag_set SAVED_MODEL_TAG_SET
Comma-separated set of tags identifying the MetaGraphDef within the SavedModel to analyze. All tags must be present. In order to pass in an empty tag set, pass in "". (default "serve")
--saved_model_signature_key SAVED_MODEL_SIGNATURE_KEY
Key identifying the SignatureDef containing inputs and outputs. (default DEFAULT_SERVING_SIGNATURE_DEF_KEY)
--enable_v1_converter
Enables the TensorFlow V1 converter in 2.0
--experimental_new_converter [EXPERIMENTAL_NEW_CONVERTER]
Experimental flag, subject to change. Enables MLIR-based conversion instead of TOCO conversion. (default True)
--experimental_new_quantizer [EXPERIMENTAL_NEW_QUANTIZER]
Experimental flag, subject to change. Enables MLIR-based quantizer instead of flatbuffer conversion. (default True)
Convert Tensorflow Model to TFlite Model
tflite_convert --saved_model_dir=YOUR_MODEL_PATH --output_file=YOUR_MODEL_NAME.tflite
Convert TFlite Model to Edge TPU Model
You should optimize your model before you convert tflite model to edge tup model, please check the Optimize Tensorflow Model
Install edgetpu compiler
curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add -
echo "deb https://packages.cloud.google.com/apt coral-edgetpu-stable main" | sudo tee /etc/apt/sources.list.d/coral-edgetpu.list
sudo apt-get update
sudo apt-get install edgetpu-compiler
Transform TFlite Model to Edge TPU Model
edgetpu_compiler YOUR_MODEL_NAME.tflite
And then you should get a new file named YOUR_MODEL_NAME_edgetpu.tflite
We do not recommend this approach because there are many conflicting packages in the actual process. And TensorFlow Lite supports a limited set of operations, some PyTorch operations may not be supported.
Convert Pytorch model to tflite model
Install dependencies
pip install -r https://github.com/google-ai-edge/ai-edge-torch/releases/download/v0.1.1/requirements.txt
pip install ai-edge-torch==0.1.1
Convert
import ai_edge_torch
import numpy
import torch
import torchvision
resnet18 = torchvision.models.resnet18(torchvision.models.ResNet18_Weights.IMAGENET1K_V1).eval()
sample_inputs = (torch.randn(1, 3, 224, 224),)
torch_output = resnet18(*sample_inputs)
edge_model = ai_edge_torch.convert(resnet18.eval(), sample_inputs)
edge_model.export('resnet.tflite')
You will get resnet.tflite
Check tflite_converter
You should optimize your model before you convert tflite model to edge tup model, please check the Optimize Tensorflow Model
tflite_convert -h
The result should be like this:
2024-07-23 10:41:03.750087: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2024-07-23 10:41:04.276520: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT
usage: tflite_convert [-h] --output_file OUTPUT_FILE [--saved_model_dir SAVED_MODEL_DIR | --keras_model_file KERAS_MODEL_FILE] [--saved_model_tag_set SAVED_MODEL_TAG_SET]
[--saved_model_signature_key SAVED_MODEL_SIGNATURE_KEY] [--enable_v1_converter] [--experimental_new_converter [EXPERIMENTAL_NEW_CONVERTER]]
[--experimental_new_quantizer [EXPERIMENTAL_NEW_QUANTIZER]]
Command line tool to run TensorFlow Lite Converter.
optional arguments:
-h, --help show this help message and exit
--output_file OUTPUT_FILE
Full filepath of the output file.
--saved_model_dir SAVED_MODEL_DIR
Full path of the directory containing the SavedModel.
--keras_model_file KERAS_MODEL_FILE
Full filepath of HDF5 file containing tf.Keras model.
--saved_model_tag_set SAVED_MODEL_TAG_SET
Comma-separated set of tags identifying the MetaGraphDef within the SavedModel to analyze. All tags must be present. In order to pass in an empty tag set, pass in "". (default "serve")
--saved_model_signature_key SAVED_MODEL_SIGNATURE_KEY
Key identifying the SignatureDef containing inputs and outputs. (default DEFAULT_SERVING_SIGNATURE_DEF_KEY)
--enable_v1_converter
Enables the TensorFlow V1 converter in 2.0
--experimental_new_converter [EXPERIMENTAL_NEW_CONVERTER]
Experimental flag, subject to change. Enables MLIR-based conversion instead of TOCO conversion. (default True)
--experimental_new_quantizer [EXPERIMENTAL_NEW_QUANTIZER]
Experimental flag, subject to change. Enables MLIR-based quantizer instead of flatbuffer conversion. (default True)
Convert TFlite Model to Edge TPU Model
Install edgetpu compiler
curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add -
echo "deb https://packages.cloud.google.com/apt coral-edgetpu-stable main" | sudo tee /etc/apt/sources.list.d/coral-edgetpu.list
sudo apt-get update
sudo apt-get install edgetpu-compiler
Transform TFlite Model to Edge TPU Model
edgetpu_compiler resnet18.tflite
And then you should get a new file named resnet18_edgetpu.tflite
Install Ultralytics
pip install ultralytics
Convert YOLO Model to egde TPU Model
# For example, if you want to convert yolov8n.pt to yolov8n_integer_quant_edgetpu.tflite
yolo export model=yolov8n.pt format=edge int8=True
The result should be like this:
jiahao@PC:~/yolov8s_saved_model$ ls
assets saved_model.pb yolov8s_float32.tflite yolov8s_full_integer_quant.tflite
fingerprint.pb variables yolov8s_full_integer_quant_edgetpu.log yolov8s_int8.tflite
metadata.yaml yolov8s_float16.tflite yolov8s_full_integer_quant_edgetpu.tflite yolov8s_integer_quant.tflite
The yolov8s_full_integer_quant_edgetpu.tflite
is the model you need.
You can convert other tflite model to edge TPU model by using the following command:
# For example, you can convert yolov8s_int8.tflite to edge TPU model
edgetpu_compiler yolov8s_int8.tflite
Tech Support & Product Discussion
Thank you for choosing our products! We are here to provide you with different support to ensure that your experience with our products is as smooth as possible. We offer several communication channels to cater to different preferences and needs.