Artificial Intelligence Learns to Describe Photos
Scientists from Stanford University developed a program called NeuralTalk that is capable of describing images’ content, writes The Verge.
The program was developed by Fei-Fei Li, a director of the Stanford Artificial Intelligence Laboratory, and Andrej Karpathy, a graduate student. NeuralTalk uses an approach similar to Google’s Deep Dream. Its archive contains a large number of the images. Each of them is described with key words. When a user downloads a photo into the program, the artificial intelligence analyzes the shapes and lines on the image and selects an analog from the archive.
After analyzing of the image, NeuralTalk provides several options of the description that allows the user to select the one that corresponds with the image the best.
{“img”: “/wp-content/uploads/2015/07/Deep-Dream_01.jpg”, “text”: “”},
{“img”: “/wp-content/uploads/2015/07/Deep-Dream_02.jpg”, “text”: “”},
{“img”: “/wp-content/uploads/2015/07/Deep-Dream_04.jpg”, “text”: “”},
{“img”: “/wp-content/uploads/2015/07/Deep-Dream_05.jpg”, “text”: “”},
{“img”: “/wp-content/uploads/2015/07/Deep-Dream_07.jpg”, “text”: “”},
{“img”: “/wp-content/uploads/2015/07/Deep-Dream_08.jpg”, “text”: “”}
NeuralTalk is still far from being perfect. For example, “a women holding a huge donut” can be identified by the program as “a girl holding a blow dryer”. The first experimental version of the program was developed last year. Recently, the scientists uploaded a demo version and provided a free access to it. You can’t upload your own photos, however you can filter the images using simple key words (cat, dog, pizza, man, woman) and examine the captions created by the artificial intelligence.
