Scene recognition is one of the branches of computer vision, which is actively used, for instance, in the search by image. The problem of scene recognition in these images is much more complicated than the object recognition problem, which is rather well-studied and already used in the industry. The main reason for it being that is that a scene is a more complex and a less formalized concept: it is quite difficult to mark out the characteristics that describe such concepts as a restaurant, a kitchen, a sport event and so forth. Besides, unlike an object, a scene takes up the whole picture, not a part of it.
In this talk we will discuss the development of a system for solving the problem of scene recognition with the help of a state-of-the-art approach based on deep convolutional neural networks.
The problem of sights recognition follows from scene recognition. Here, out of all scenes, we need to tell apart those which include sights: palaces, monuments, squares, cathedrals and so on. In the mean time, it is also important to keep the false detection rate negligible. In this talk we will consider the sights recognition problem solving based on scene recognition neural networks.