What is the first thing that comes to my mind when I want to curse somebody? May your CAPTCHAs always and forever be an unreadable mess! The best of CAPTCHAs make you question your humanity, and of bots! The sole reason why you have to fill them out is to prevent automated logins by bots. But with progress in machine learning and image recognition, computer vision has eventually reached a point where it may turn out to be quite useful to us.
Flickr, photo/video hosting site recently launched Park or Bird. The single-page web app was built-in response to an XKDC comic lampooning at the limitations of computers that are still blind as a bat when it comes to perception image content. Park or Bird allows users to upload an image and automatically determine whether the image taken is of a national park or of a bird.
Flickr’s Image Recognition Technology
Flicker’s image recognition technology utilizes deep convolutional neural networks. The influential advancement in computer vision will soon be integrated in Flickr’s product roadmap and in the near future may even become an integral part of our daily lives. Google too, has detailed its image recognition technology by investing in deep learning technique. In 2014, it acquired two image recognition startups, Jetpac and DNNResearch, to take its image recognition capabilities on its photo app to the next level.
Simon Osindero, an AI architect in the Flicker Vision and Machine Learning group affirmed that with “key algorithmic improvements and the availability of more powerful computing infrastructures”, their deep learning algorithm works well “particularly for object, scene and attribute recognition in photos”. It can recognize over 1,000 different objects in images. This is done by passing images through a string of layers. Each of these layers transforms the original image and executes progressively more and more complex computations. Additionally Flicker also utilizes open source tools such as the deep learning framework from UC Berkeley commonly known as Caffe.
Recognize 1,000 Different Objects In An Image
Flickr already has some very advanced technology in their photo search, where users can easily search through their rapidly expanding portfolio of images. The new image recognition technology can turn out to be immensely beneficial. With our ever-growing collection of images on cloud, keeping them organized and easily accessible will turn more and more complex. In fact, this is what an ex-Apple employee Tim Bucher’s startup Lyve does.
Presently Flickr is working on improvements essential for balancing computer knowledge with human annotations. Although a challenge, the Yahoo-owned company is keen on improving its technology over time.