Yahoo Open Sources Porn-Hunting Neural Network
Source: Stephanie Mlot
Yahoo has open-sourced its deep learning network, which can be used to classify pornographic content on the Web.
The artificial intelligence system is trained to automatically identify risque images using a probability scale between zero and one. Scores below 0.2 indicate the image is likely safe for all eyes. But those above 0.8 signal the high probability of a long chat with your boss if they spot your computer screen.
A picture of a woman covered from the neck down and standing in a field of flowers, for instance, earns a score of 0.001—about as inoffensive as it gets. A shirtless man walking on the beach, however, might draw a score of 0.074, while two bikini-clad female lifeguards running in the sand scores 0.116.
"Defining NSFW materials is subjective and the task of identifying these images is non-trivial," Yahoo research engineer Jay Mahadeokar and senior director of product management Gerry Pesavento wrote in a blog post. "Moreover, what may be objectionable in one context can be suitable in another."
Inappropriate sketches, cartoons, text, images of graphic violence, and other content are not currently addressed.
The new model, trained using the CaffeOnSpark system, works with the general-purpose Caffe open-source deep learning framework. Developers may have to fine-tune the program to fit their needs.
"Since images and user-generated content dominate the Internet today, filtering NSFW [not suitable for work] images becomes an essential component of Web and mobile applications," the blog said.
Facebook, Google, and Microsoft have open-sourced different deep learning systems in the past. But Yahoo's program is still in its infancy.
"The definition of NSFW is subjective and contextual. This model is a general purpose reference model, which can be used for the preliminary filtering of pornographic images," the open nsfw project authors wrote in a disclaimer. "We do not provide guarantees of accuracy of output, rather we make this available for developers to explore and enhance as an open-source project."
| }
|