Commoditizing deep learning is mandatory. After repetitive in production installs at various corps while connecting to existing pipelines, I ve convinced some o

Commoditizing deep learning is mandatory. After repetitive in production installs at various corps while connecting to existing pipelines, I ve convinced some of them to sponsor a commoditized open source deep learning server. Code is here: https://github.com/beniz/deepdetect https://github.com/beniz/deepdetect There are differenciated CPU and GPU docker versions, and as mentioned elsewhere in this thread, they are the easiest way to setup even production system without critical impact on performances, thanks to nvidia-docker. It seems they are more popular than AMI within our little community.