Incremental Learning for Semantic Segmentation

Deep learning architectures exhibit a critical drop of performance due to catastrophic forgetting when they are required to incrementally learn new tasks. Incremental learning for semantic segmentation is the ability of a learning system (e.g., a neural network) to learn the segmentation and the labeling of the new classes without forgetting or deteriorating too much the performance on previously learned ones. The target is to develop novel incremental learning techniques based on deep learning for semantic segmentation.

Unsupervised Domain Adaptation for Semantic Segmentation

Unsupervised domain adaptation for semantic segmentation is the task of aligning a network trained on source data to perform well on target data. Complex deep neural networks for this task require to be trained with a huge amount of labeled data, which is difficult and expensive to acquire. A recently proposed workaround is the usage of synthetic data, however the differences between real world and synthetic scenes limit the performance. The target is to develop novel unsupervised domain adaptation techniques for deep networks.

3D Data Acquisition and Fusion from Multiple sensors

The target is the acquisition of highly accurate static and dynamic 3D representations by combining the data coming from multiple sensors, including stereo vision systems, time-of-flight cameras and structured light devices. Various fusion algorithms will be exploited including techniques based on deep learning eventually with domain adaptation strategies.

Hand Gesture Recognition

The target is the recognition of static and dynamic gestures of the hand. For this task 3D information acquired by depth cameras will be used together with color images. Different machine learning techniques will also be exploited.

SONY

 Stages are available at the SONY STC research center in Stuttgart (Germany). 

NIDEK TECHNOLOGIES

Stages are available at Nidek Technologies on gesture-based interfaces and other topics.