This is a modern environment for building deep learning applications. It has the latest stable versions of the most common tools and frameworks that you're likely to need.
Keep in mind that this image is big (3GB+). I considered dropping a few tools or creating different images with different toolsets, but I think that'll waste everyone's time. If you're doing deep learning then you probably have a lot of disk space anyway, and you're likely to prefer saving time over disk space.
- Ubuntu 16.04 LTS
- Python 3.5.2
- Tensorflow 1.6.0
- Keras 2.1.5
- PyTorch 0.3.1
- OpenCV 3.4.1
- Jupyter Notebook
- Numpy, Scipy, Scikit Learn, Scikit Image, Pandas, Matplotlib, Pillow
- Caffe
- Java JDK
- PyCocoTools (MS COCO dev kit)
TODO:
- GPU/CUDA (due to Docker hub time limits, auto builds fail to build this. Suggestions welcome)
If you need to run older models that require Python 2.7 or OpenCV 2.4 then I'd recommend Sai's docker image.
If you haven't yet, start by installing Docker. Then run this command at your terminal and it will open a bash prompt inside the container.
docker run -it -p 8888:8888 -p 6006:6006 -v ~/:/host waleedka/modern-deep-learning
Note the -v option. It maps your user directory (~/) to /host in the container. Change it if needed. The two -p options expose the ports used by Jupyter Notebook and Tensorboard respectively.
Important: Do not run this on a public server accessible from the Internet. Security features have been disabled in the settings for convenience of local development.
While inside the Docker container (see previous section) run this command, then navigate to: http://localhost:8888/
cd /host # So Jupyter Notebook uses this as it's root
jupyter notebook --allow-rootAlternatively, combine the previous two steps and start Jupyter Notebook without logging into the container:
docker run -it -p 8888:8888 -p 6006:6006 -v ~/:/host waleedka/modern-deep-learning jupyter notebook --allow-root /hostSubmit issues and pull requests in the GitHub repo.