Face R-FCN算法笔记

论文：Detecting Faces Using Region-based Fully Convolutional Networks
链接：https://arxiv.org/abs/1709.05256

腾讯AI Lab的文章，总结起来就是在R-FCN框架上做了一些修改，使其更好地应用在人脸检测上，修改主要包含三个方面，可以看原文的这段话：We improve the R-FCN framework for targeting face detection in three aspects. First, we introduce additional smaller anchors and modify the position sensitive RoI pooling to a smaller size for suiting the detection of the tiny faces. Second, we propose to use position-sensitive average pooling instead of normal average pooling for the last feature voting in R-FCN, which leads to an improved embedding. Third, multi-scale training strategy and on-line Hard Example Mining (OHEM) strategy are adopted for training. In the testing phase, we also ensemble the multi-scale detection results to improve the performance. 可以看出主要是一些trick，比如用更小尺寸的anchor、将原来R-FCN中最后的vote操作由average p