Real-Time Image Recognition for an Android App

computer vision

Client

An Android application developer. Market leaders in file recovery tools for Android. They were working on an app to help users filter and hide explicit content on their devices.

Problem

The client tried several solutions, including Open-Source AI Software, which did not perform well and failed to deliver expected results. So they approached Broutonlab to develop a model for detecting explicit visual content in real-time.

Solution

Data Scientists at BroutonLab trained an artificial neural network for mobile platforms that filters NSFW content in real-time.


The neural network works with low-quality and high-quality images and videos and has several sensitivity settings. The settings of the network determine what content is considered NSFW.


The network is small and fast enough to run on mobile phones (including phones without GPUs), and more accurate than Google Vision and Microsoft Cognitive Services.

The neural network consists of several convolutional layers that extract features from the input, and a classification head that makes the final decision. It classifies videos based on the output for their frames.

Get a Free Consultation Now