X-LINUX-AI - image classification using Coral Edge TPU TensorFlow Lite C++ API

This article explains how to experiment with Coral Edge TPU[1] applications for image classification based on the MobileNet v1 model using the TensorFLow Lite C++ API

1 Description[edit]

The image classification[2] neural network model allows identification of the subject represented by an image. It classifies an image into various classes.

C/C++ Coral Edge TPU image classification application

The application demonstrates a computer vision use case for image classification where frames are grabbed from a camera input (/dev/videox) and analyzed by a neural network model executed on the Coral Edge TPU[1] using the TensorFlow Lite C++ API.
A Gstreamer pipeline is used to stream camera frames (using v4l2src), to display a preview (using waylandsink) and to execute neural network inference (using appsink).
The result of the inference is displayed in the preview. The overlay is done using GtkWidget with cairo.
This combination is quite simple and efficient in terms of CPU overhead.

The model used with this application is the MobileNet v1 downloaded from the Coral GitHub testing models[3].

2 Installation[edit]

2.1 Install from the OpenSTLinux AI package repository[edit]

Warning.png The software package is provided AS IS, and by downloading it, you agree to be bound to the terms of the software license agreement (SLA). The detailed content licenses can be found here.

After having configured the AI OpenSTLinux package you can install X-LINUX-AI components for this application:

Board $> apt-get install tflite-cv-apps-edgetpu-image-classification-c++

And restart the demo launcher:

Board $> systemctl restart weston@root

2.2 Source code location[edit]

  • in the Openembedded OpenSTLinux Distribution with X-LINUX-AI Expansion Package:
<Distribution Package installation directory>/layers/meta-st/meta-st-stm32mpu-ai/recipes-samples/tflite-cv-apps-edgetpu/files/image-classification/src
  • on GitHub:
https://github.com/STMicroelectronics/meta-st-stm32mpu-ai/tree/dunfell/recipes-samples/tflite-cv-apps-edgetpu/files/image-classification/src

2.3 Re-generate the package from OpenSTLinux Distribution (optional)[edit]

Using the Openembedded OpenSTLinux Distribution, you are able to rebuild the application.

Info.png If not already installed, the X-LINUX-AI OpenSTLinux distribution need to be installed by following this link


  • Set up the build environment:
PC $> cd <Distribution Package installation directory>
PC $> source layers/meta-st/scripts/envsetup.sh
  • Rebuild the application:
PC $> bitbake tflite-cv-apps-edgetpu-image-classification-c++ -c compile

The generated binary is available here:

<Distribution Package installation directory>/<build directory>/tmp-glibc/work/cortexa7t2hf-neon-vfpv4-ostl-linux-gnueabi/tflite-cv-apps-edgetpu-image-classification-c++/1.0-r0/tflite-cv-apps-edgetpu-image-classification-c++-1.0/image-classification/src

3 How to use the application[edit]

3.1 Launching via the demo launcher[edit]

Launch cpp tfl edgetpu image classification.png

3.2 Executing with the command line[edit]

The label_tfl_edgetpu_gst_gtk C/C++ application is located in the userfs partition:

/usr/local/demo-ai/computer-vision/tflite-image-classification-edgetpu/bin/label_tfl_edgetpu_gst_gtk

It accepts the following input parameters:

Usage: ./label_tfl_edgetpu_gst_gtk -m <model .tflite> -l <label .txt file>                                                            
                                                                                                                                      
-m --model_file <.tflite file path>:  .tflite model to be executed                                                                    
-l --label_file <label file path>:    name of file containing labels                                                                  
-i --image <directory path>:          image directory with image to be classified                                                     
-v --video_device <n>:                video device (default /dev/video0)                                                              
--crop:                               if set, the nn input image is cropped (with the expected nn aspect ratio) before being resized, 
                                      else the nn imput image is only resized to the nn input size (could cause picture deformation). 
--frame_width  <val>:                 width of the camera frame (default is 640)                                                      
--frame_height <val>:                 height of the camera frame (default is 480)                                                     
--framerate <val>:                    framerate of the camera (default is 15fps)                                                      
--input_mean <val>:                   model input mean (default is 127.5)                                                             
--input_std  <val>:                   model input standard deviation (default is 127.5)                                               
--help:                               show this help

3.3 Testing with MobileNet V1[edit]

The model used for test is the mobilenet_v1_1.0_224_quant_edgetpu.tflite downloaded from Coral GitHub testing models[3].

Info.png The different objects the neural network is able to classify are listed in the labels.txt file located in the target:

/usr/local/demo-ai/computer-vision/models/mobilenet/labels.txt


To ease launching of the application, two shell scripts are available:

  • launch image classification based on camera frame inputs
Board $> /usr/local/demo-ai/computer-vision/tflite-image-classification-edgetpu/bin/launch_bin_label_tfl_edgetpu_mobilenet.sh
  • launch image classification based on the pictures located in /usr/local/demo-ai/computer-vision/models/mobilenet/testdata directory
Board $> /usr/local/demo-ai/computer-vision/tflite-image-classification-edgetpu/bin/launch_bin_label_tfl_edgetpu_mobilenet_testdata.sh
Info.png Note that you need to populate the testdata directory with your own data sets.

The pictures are then randomly read from the testdata directory

4 References[edit]