Leveraging deep neural networks to fight child pornography in the age of social media.

•Our totally data-driven method distinguish SEIC from adult and innocuous content.•Our solution gives an error reduction of up to 70% comparing to skin detector systems.•In the case of BoVW-based method, our solution leads to an error reduction of 56%.•Our solution shows remarkable improvements over also forensic and commercial tools.•We produce a virtual machine, ready for initial deployments by law-enforcement agents.