Learning to count.
Learning to count is a learning strategy that has been recently proposed in the literature for dealing with problems where estimating the number of object instances in a scene is the final objective. In this framework, the task of learning to detect and localize individual object instances is seen as a harder task that can be evaded by casting the problem as that of computing a regression value from hand-crafted image features. In this project we explore the features that are learned when training a counting convolutional neural network in order to understand their underlying representation.
The main hypothesis of this line is that the multiplicity of the instances of the concept of interest in an image provides strong information regarding their discriminability for a feature learning process to exploit. CNN provide a nice framework for this problem since they naturally handle feature learning and have shown impressive classification performance on different benchmark problems.
A second interesting question we could ask is how well these descriptors represent individual instances, i.e. does the network learn a representation for instance recognition?