The example code that shows how to load the Google Cloud AutoML Video Object Tracking On-Device models and conduct inference on a sequence of images from a video clip.
The targeted devices are CPU and Edge TPU.
- Yongzhe Wang (yongzhe@google.com)
- Henry Quoc Tran (henryquoctran@google.com)
Launch a docker shell.
make -f ondevice-examples.makefile docker-example-shell
Enter working directory.
cd /edgetpu-ml-cpp
Build the binaries. For development, stay in the docker shell and re-run the following command to create a new build.
make -f ondevice-examples.makefile ondevice-examples
The resulting binaries will be copied to cpp_example_out/ondevice_demo_*.
The ondevice-examples command will build binaries for amd64 (desktop), arm64, and arm32.
For faster development, use a platform specific makefile rule:
make -f ondevice-examples.makefile ondevice-examples-arm64
Running ./ondevice_demo will create .txt files with the classes, score, and bounding boxes. To visualize the output, you may run the following command while still in the docker shell:
bazel run //tools:visualizer -- --image_path=`pwd`/output/00001.bmp --result_path=`pwd`/output/00001.bmp.txt --output_path=`pwd`/00001_visualized.bmp
Note that bazel does not execute from the working directory, so the paths are prefixed with `pwd`. This is not necessary if an absolute path is given.