PROJECT TITLE :

Region-Based Convolutional Networks for Accurate Object Detection and Segmentation

ABSTRACT:

Object detection performance, as measured on the canonical PASCAL VOC Challenge datasets, plateaued in the ultimate years of the competition. The most effective-performing methods were advanced ensemble systems that usually combined multiple low-level image features with high-level context. During this paper, we propose a straightforward and scalable detection algorithm that improves mean average precision (mAP) by a lot of than fifty percent relative to the previous best result on VOC 2012—achieving a mAP of sixty two.4 %. Our approach combines 2 ideas: (1) one can apply high-capacity convolutional networks (CNNs) to bottom-up region proposals in order to localize and segment objects and (two) when labeled coaching knowledge are scarce, supervised pre-coaching for an auxiliary task, followed by domain-specific fine-tuning, boosts performance considerably. Since we mix region proposals with CNNs, we tend to decision the resulting model an R-CNN or Region-based mostly Convolutional Network. Supply code for the complete system is out there at http://www.cs.berkeley.edu/~rbg/rcnn.