Deep learning made easy
I have been experimenting with AI and deep learning for some years now and wanted to make it easier for developers to do a project with tensorflow and be up and running quickly. Before I was struggling with installing CUDA, tensorflow and python on my system, and keeping them up to date with correct versions which was time consuming. I solved it by process isolation in a docker container that has everything including
- NVIDIA docker
- CUDA
- Python 3
- Tensorflow-gpu
- Ubuntu 18.04
- Tools
All you need is a PC with an NVIDIA GPU or even better an AWS instance for deep learning. It is now easier than ever to start doing object detection for example. Just follow the TRAINING.md instructions in my github repo which will guide you through the steps to get started. I hope you can be more successful in your next deep learning project using tensorflow and docker.
Example to get started (navigate to the root github repo cloned by git clone):
make jupyter
And navigate to
./models/research/object_detection/object_detection_tutorial.ipynb
And run the first object detection example. If you need to run on your web camera or a video stream start using the python script under scripts/:
python3 test_object_detection.py