Are you looking for something to keep unsafe content away from the office or at home (from the kids as well as the husband)? Yahoo’s new open source artificial intelligence can help you with this. Jay Mahadeokar – Yahoo’s research engineer, and Gerry Pesavento – senior director of product management, recently published on Friday announcing the release of the company’s “deep learning model that will allow developers to experiment with a classifier for NSFW detection, and provide feedback to Yahoo on ways to improve the classifier.”
In short, Yahoo is open sourcing its algorithms for detecting pornographic images. The automatic identification of any image that is NSFW (Not Suitable/Safe For Work) including offensive and adult images, has been a long standing problem that researchers have been trying so solve for decades. With the technological advancement of computer vision, improved training data, and deep learning algorithms, computers can now automatically classify NSFW image content with greater precision.
However, an open source algorithm model for identifying images for NSFW does not currently exist. This is where Yahoo comes in to fill in the gap. It is also hoped that this learning tool will be used by programmers to identify pornographic images on work desktops. Of course, while what may be objectionable in one context may not be suitable in another, but on the whole, any form of pornography is definitely unsuitable in all contexts.
Yahoo’s open source deep learning tool system can assign an image a score between 0 and 1 to determine its NSFW value. It can also be used to score an image in ranking search engine results. However, the accuracy of the deep learning tool will fully depend on the programmer or developer. The preliminary filtering of pornographic images will generally depend on the general purpose reference model. In short, the guaranteed accuracy output of the learning tool will depend on how the user trains the tool. Thus, developers are encouraged to explore and enhance the open source project.
Video Courtesy of YouTube:
Remember that while the human eye is automatically trained by experience to identify what is porn, machines are completely innocent. The open source project can be trained as an image-recognition engine using literally thousands of pornographic pictures. This is also to train the NSFW detector that normal online searches are less likely to contain the intimate parts of the human body, even at random. Databases must also contain other specific types of imagery for training on certain patterns.
For instance, cars are identified by recognizing wheels, door handles, etc., dogs by their tails, snouts, fur, etc., and people with clothes by cloth color, patterns, style, etc. So if its porn, well, you can use your imagination, and the absence of clothes. But again, the programmers and developers need to provide their own porn to train the model. There are more details available at the Yahoo blog post where the article was first published, and the open source model is available for download on GitHub.