/docker-frcnn-tennis

Dockerfile for building and running https://github.com/anujkhare/py-faster-rcnn/tree/noplot

Requirements

Instructions to train

This build works for computers with a CUDA-enabled GPU. It is possible to build for CPU as well, but this Dockerfile currently does not support it.

  1. Clone the repository and navigate to it.
  2. Edit Dockerfile. GPU_ARCH needs to set manually as CUDA does not work during the build.
    GPU_ARCH=<GPU-Arch>

  1. Build the docker image using: docker build -t anujkhare/rcnn:cudnn5 . Note: This step may take a while.

  2. Download the data into bin/data.

  3. (Optional) Verify the build by running a container:

nvidia-docker run -it -v /path/to/bin/data:/opt/code/frcnn/py-faster-rcnn/data anujkhare/rcnn:cudnn5

Inside the container:

cd /opt/code/frcnn/py-faster-rcnn/caffe-fast-rcnn/build
make runtest
  1. Run the container with VNC server for GUI:
nvidia-docker run --name racket-train -e HOME=/ -p 5900 -v /path/to/bin/data:/opt/code/frcnn/py-faster-rcnn/data anujkhare/rcnn:cudnn5 x11vnc -forever -usepw -create
  1. Connect to the VNC server Find the host port on which VNC is running using:
docker ps

E.g., if PORTS, the host port

Note: In TigerVNC viewer, pressing F8 opens the context menu. Very important to know if you go into full-screen in the VNC viewer, since all the keys are captured by it!