Deep learning architectures exhibit a critical drop of performance due to catastrophic forgetting when they are required to incrementally learn new tasks. Contemporary incremental learning frameworks focus on image classification and object detection while in this work we formally introduce the incremental learning problem for semantic segmentation in which a pixel-wise labeling is considered. To tackle this task we propose to distill the knowledge of the previous model to retain the information about previously learned classes, whilst updating the current model to learn the new ones. We propose various approaches working both on the output logits and on intermediate features. In opposition to some recent frameworks, we do not store any image from previously learned classes and only the last model is needed to preserve high accuracy on these classes. The experimental evaluation on the Pascal VOC2012 dataset shows the effectiveness of the proposed approaches.

The recent introduction of novel acquisition devices like the Leap Motion and the Kinect allows to obtain a very informative description of the hand pose that can be exploited for accurate gesture recognition. This paper proposes a novel hand gesture recognition scheme explicitly targeted to Leap Motion data. An ad-hoc feature set based on the positions and orientation of the fingertips is computed and fed into a multi-class SVM classifier in order to recognize the performed gestures. A set of features is also extracted from the depth computed from the Kinect and combined with the Leap Motion ones in order to improve the recognition performance. Experimental results present a comparison between the accuracy that can be obtained from the two devices on a subset of the American Manual Alphabet and show how, by combining the two features sets, it is possible to achieve a very high accuracy in real-time

No abstract

No abstract

Novel 3D acquisition devices like depth cameras and the Leap Motion have recently reached the market. Depth cameras allow to obtain a complete 3D description of the framed scene while the Leap Motion sensor is a device explicitly targeted for hand gesture recognition and provides only a limited set of relevant points. This paper shows how to jointly exploit the two types of sensors for accurate gesture recognition. An ad-hoc solution for the joint calibration of the two devices is firstly presented. Then a set of novel feature descriptors is introduced both for the Leap Motion and for depth data. Various schemes based on the distances of the hand samples from the centroid, on the curvature of the hand contour and on the convex hull of the hand shape are employed and the use of Leap Motion data to aid feature extraction is also considered. The proposed feature sets are fed to two different classifiers, one based on multi-class SVMs and one exploiting Random Forests. Different feature selection algorithms have also been tested in order to reduce the complexity of the approach. Experimental results show that a very high accuracy can be obtained from the proposed method. The current implementation is also able to run in real-time