Deep Learning for Image Recognition

Abstract

With recent rapid advances in deep learning, there have been major breakthroughs in computer vision, one of the core subareas of artificial intelligence. In this talk, I present representative computer vision works from my research group with a focus on object recognition and scene understanding. Specifically, I present deep learning algorithms for image classification, RGBD scene labeling as well as salient object detection. For image classification, I introduce a deep transfer learning scheme, called selective joint fine-tuning, for boosting the performance of tasks with insufficient training data. In this scheme, a target learning task with insufficient training data is carried out simultaneously with another source learning task with abundant training data. However, the source learning task only uses a subset of training data most relevant to the target learning task. Experiments demonstrate that our deep transfer learning scheme achieves state-of-the-art performance on multiple visual classification tasks with insufficient training data for deep learning.