This is a docker container for https://github.com/yosinski/deep-visualization-toolbox to get the toolbox running with CPU only without going through the hassle of manual installation.
Please check out https://github.com/fishcorn/dvtb-container for a docker container of the same toolbox but runs with CUDA GPUs. I created this container because fishcorn/dvtb-container did not work on my MacBook which lacks Nvidia GPU.
First, pull the image from the docker-hub using:
docker pull mhalsharif/dvtb-cpu
To enable X server connection between the host and the container, run:
xhost +
On Linux (video cam input is supported), run this container with:
docker run -ti --name dvtb-container --device=/dev/video0:/dev/video0 --env="DISPLAY" --net=host -v /tmp/.X11-unix:/tmp/.X11-unix mhalsharif/dvtb-cpu
On macOS (video cam input is not supported), run this container with:
docker run -ti --name dvtb-container -e DISPLAY=host.docker.internal:0 -v /tmp/.X11-unix:/tmp/.X11-unix mhalsharif/dvtb-cpu
Before running the toolbox, make sure that you have downloaded (fetched) one of the models in the 'models' folder. For instance:
cd models
cd caffenet-yos
./fetch.sh
Finally, run the toolbox in /opt/dvtb folder with:
./run_toolbox.py
First clone this repository:
git clone https://github.com/mhalsharif/dvtb-cpu.git
Then build the image with:
docker build -t dvtb-cpu .