CVFeb 7, 2017

An Implementation of Faster RCNN with Study for Region Sampling

arXiv:1702.02138v2166 citations
AI Analysis

This is an incremental study providing a public implementation and analysis of region sampling for object detection.

The researchers implemented Faster RCNN in TensorFlow and studied region sampling methods, finding that biased sampling toward small regions achieves comparable mean average precision (mAP) to non-maximal suppression-based sampling when sufficiently converged.

We adapted the join-training scheme of Faster RCNN framework from Caffe to TensorFlow as a baseline implementation for object detection. Our code is made publicly available. This report documents the simplifications made to the original pipeline, with justifications from ablation analysis on both PASCAL VOC 2007 and COCO 2014. We further investigated the role of non-maximal suppression (NMS) in selecting regions-of-interest (RoIs) for region classification, and found that a biased sampling toward small regions helps performance and can achieve on-par mAP to NMS-based sampling when converged sufficiently.

Code Implementations47 repos

Data from Papers with Code (CC-BY-SA-4.0)

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes