Public Repository

Last pushed: 2 years ago
Short Description
This image trained a neural network to demonstrate the *Tunnel Effect in CNNs (2016)*
Full Description

Tunnel Effect

This image trained a neural network to demonstrate the Tunnel Effect in CNNs. It was built to run on GPU with Nut / nvidia-docker.

How To Use

1) Run with nvidia-docker:

nvidia-docker run -it --rm --workdir=/nut/tunneleffect/ --volume=/folder/of/your/choice:/nut/tunneleffect  --env="NUT_VOLUME_MAIN_PATH=/nut/tunneleffect/" --env="NUT_VOLUME_DATASET_PATH=/nut/tunneleffect/dataset"  matthieudelaro/tunneleffect64

With this command, we tell docker to run the image with two environment variables. We also give access to the GPU(s) of the host (nvidia-docker handles this). We also share the /folder/of/your/choice with the container, so that it can save the weights, the logs, and the reconstructed images on the host. Upon completion of the training, the container will be destroyed. But all the files created by the container will persist in the /folder/of/your/choice.

2) Or rely on nvidia-docker's driver, but specifying all parameters by yourself:

docker run -it --rm --workdir=/nut/tunneleffect/  --volume=/folder/of/your/choice:/nut/tunneleffect   --volume-driver=nvidia-docker --volume=/var/lib/docker/volumes/nvidia_driver_352.63/_data:/usr/local/nvidia:ro --env="NUT_VOLUME_MAIN_PATH=/nut/tunneleffect/" --env="NUT_VOLUME_DATASET_PATH=/nut/tunneleffect/dataset" --device="/dev/nvidia0:/dev/nvidia0:mrw" --device="/dev/nvidiactl:/dev/nvidiactl:mrw" --device="/dev/nvidia-uvm:/dev/nvidia-uvm:mrw" matthieudelaro/tunneleffect64

If you need to adapt this command to your configuration, here is how is was built:

nut --dockercli --macro=tunnel --logs
# => copy the full command from the logs
# => change the docker image name
# In case you encounter the following error message, "docker: Error response from daemon: create nvidia_driver_352.63: conflict: volume name must be unique."
# replace the volume name with its path (Mountpoint given by `docker volume inspect nvidia_driver_352.63`)
Docker Pull Command