vix.ing · top · new · best · stats · spec

Rich feature hierarchies for accurate object detection and semantic\n segmentation

2013/11/11 by Ross Girshick, Jeff Donahue, Girshick, Ross +5 · 242 citations
Computer Science · #Advanced Image and Video Retrieval Techniques #Advanced Neural Network Applications #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Multimodal Machine Learning Applications

paper · pdf · doi:10.48550/arxiv.1311.2524

openalex publication_date 2013/11/11 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Object detection performance, as measured on the canonical PASCAL VOC\ndataset, has plateaued in the last few years. The best-performing methods are\ncomplex ensemble systems that typically combine multiple low-level image\nfeatures with high-level context. In this paper, we propose a simple and\nscalable detection algorithm that improves mean average precision (mAP) by more\nthan 30% relative to the previous best result on VOC 2012---achieving a mAP of\n53.3%. Our approach combines two key insights: (1) one can apply high-capacity\nconvolutional neural networks (CNNs) to bottom-up region proposals in order to\nlocalize and segment objects and (2) when labeled training data is scarce,\nsupervised pre-training for an auxiliary task, followed by domain-specific\nfine-tuning, yields a significant performance boost. Since we combine region\nproposals with CNNs, we call our method R-CNN: Regions with CNN features. We\nalso compare R-CNN to OverFeat, a recently proposed sliding-window detector\nbased on a similar CNN architecture. We find that R-CNN outperforms OverFeat by\na large margin on the 200-class ILSVRC2013 detection dataset. Source code for\nthe complete system is available at http://www.cs.berkeley.edu/~rbg/rcnn.\n

Citations

Cited by

Related