Bumble, developing one of the largest acquaintances of dating, Opened the initial texts of the machine learning system private detector used to determine obscene images in the services uploaded to the service. . The system is written in Python, uses Tensorflow Freimvork and spreads under the Apache-2.0 license. For the classification, a reserve neural network is used efficientnet v2 . To download available The finished model to identify images of naked people. The accuracy of the definition is more
98%.
The composition also includes a script to create your own models that can be trained in your collection and used to classify arbitrary contents. For learning, just start a script with text files containing images lists with positive and negative properties. After completing the training, you can transfer an arbitrary image in Private Detector and on its basis will be calculated the entry coefficient.