R-FCN:Paper
lastmod 2026-11-10created 2016-11-022 edits
R-FCN (Region-based Fully Convolutional Networks) is a paper that proposes a deep learning model performing region-based object detection with a fully convolutional architecture. Unlike Faster R-CNN, which runs a subnetwork repeatedly for each region proposal, it computes position-sensitive score maps once over the whole image and handles classification and box regression with RoI pooling, which improves speed. This page summarizes the contents of that paper.