Rich feature hierarchies for accurate object detection and semantic segmentation

TL;DR

This paper introduces R-CNN, a simple yet effective deep learning-based object detection method that significantly outperforms previous approaches by combining region proposals with CNN features and leveraging pre-training.

Contribution

The paper presents R-CNN, a novel approach that applies high-capacity CNNs to region proposals, improving object detection accuracy and demonstrating the benefits of pre-training and fine-tuning.

Findings

R-CNN achieves over 30% improvement in mAP on VOC 2012.

R-CNN outperforms OverFeat on the ILSVRC2013 dataset.

Pre-training plus fine-tuning significantly boosts detection performance.

Abstract

Object detection performance, as measured on the canonical PASCAL VOC dataset, has plateaued in the last few years. The best-performing methods are complex ensemble systems that typically combine multiple low-level image features with high-level context. In this paper, we propose a simple and scalable detection algorithm that improves mean average precision (mAP) by more than 30% relative to the previous best result on VOC 2012---achieving a mAP of 53.3%. Our approach combines two key insights: (1) one can apply high-capacity convolutional neural networks (CNNs) to bottom-up region proposals in order to localize and segment objects and (2) when labeled training data is scarce, supervised pre-training for an auxiliary task, followed by domain-specific fine-tuning, yields a significant performance boost. Since we combine region proposals with CNNs, we call our method R-CNN: Regions with CNN features. We also compare R-CNN to OverFeat, a recently proposed sliding-window detector based on a similar CNN architecture. We find that R-CNN outperforms OverFeat by a large margin on the 200-class ILSVRC2013 detection dataset. Source code for the complete system is available at http://www.cs.berkeley.edu/~rbg/rcnn.