Deep learning for human part discovery in images

This paper addresses the problem of human body part segmentation in conventional RGB images, which has several applications in robotics, such as learning from demonstration and human-robot handovers. The proposed solution is based on Convolutional Neural Networks (CNNs). We present a network architecture that assigns each pixel to one of a predefined set of human body part classes, such as head, torso, arms, legs. After initializing weights with a very deep convolutional network for image classification, the network can be trained end-to-end and yields precise class predictions at the original input resolution. Our architecture particularly improves on over-fitting issues in the up-convolutional part of the network. Relying only on RGB rather than RGB-D images also allows us to apply the approach outdoors. The network achieves state-of-the-art performance on the PASCAL Parts dataset. Moreover, we introduce two new part segmentation datasets, the Freiburg sitting people dataset and the Freiburg people in disaster dataset. We also present results obtained with a ground robot and an unmanned aerial vehicle.

Authors

• Gabriel L. Oliveira
• Abhinav Valada
• Claas Bollen
• Wolfram Burgard
• Thomas Brox

Keywords

• Image segmentation
• Robots
• Semantics
• Image resolution
• Training
• Proposals
• Feature extraction
• Deep Learning
• Convolutional Network
• Convolutional Neural Network
• Body Parts
• Image Classification
• Part Of Network
• Unmanned Aerial Vehicles
• Part Segmentation
• Parts Of The Human Body
• Ground Robots
• Training Data
• Support Vector Machine
• Convolutional Layers
• Feature Maps
• Data Augmentation
• Class Labels
• Intersection Over Union
• Stochastic Gradient Descent
• Semantic Segmentation
• Pose Estimation
• Fully Convolutional Network
• Conditional Random Field
• Pixel Accuracy
• Type Of Augmentation
• Real Robot
• Region Proposal
• Augmentation Strategy
• Semantic Segmentation Task
• Convolutional Neural Network Features
• Level Of Granularity