Speeding-up Object Detection Training for Robotics with FALKON
share
Latest deep learning methods for object detection provided remarkable performance boost, but have limits when used in robotic applications. One of the most relevant issues is the long training time, which is due to the large size and unbalance of the associated training sets, characterized by few positive and tons of negative (i.e. background) examples. Proposed approaches, either based on end-to-end learning by back-propagation [22], or standard kernel methods trained with Hard Negatives Mining on top of deep features [8], proved to be effective, but prohibitively slow for on-line applications. In this paper we propose a novel pipeline for object detection that overcomes this problem and provides comparable performance, with a 60x training speedup. Our pipeline combines (i) the Region Proposal Network and the deep feature extractor from [22] to efficiently select candidate RoIs and encode them into powerful representations, with (ii) the recently proposed FALKON [23] algorithm, a novel kernel-based method that allows to quickly train on millions of points. We address the size and unbalance of training data by exploiting the stochastic subsampling intrinsic into the method, combined with a novel, fast, bootstrapping approach. We assess the effectiveness of the approach in a standard computer vision setting (PASCAL VOC 2007 [5]) and demonstrate its applicability to a real robotic scenario as represented by the iCubWorld Transformations [18] dataset.
READ FULL TEXT