unsurv.org

exposing surveillance

Train your object detection model for your coral dev board with your GPU

de / en

10 Apr, 2021

You can easily use your GPU to train object detection models for your coral dev board (mini). Here is how:

Follow the instructions to pass NVIDIA GPUs to tensorflow docker containers. Using this https://coral.ai/docs/edgetpu/retrain-detection/#set-up-the-docker-container stop before building the docker container.

In the ".../tutorials/docker/object_detection/Dockerfile" edit the following line

FROM tensorflow/tensorflow:1.12.0-rc2-devel

to

FROM tensorflow/tensorflow:1.12.0-rc2-devel-gpu

Follow the tutorial to build the container. When using docker-run add "--gpus all" to your command. You should be able to use the nvidia-smi command from inside your container now. It should start like this:

sudo docker run --name edgetpu-detect --gpus all --rm -it --privileged -p 6006:6006

Follow the rest of the tutorial from coral. That's it, have fun.