m0_37223909的博客

1. 目标检测之YOLO V1[2015, Joseph]1.1. YOLO V1: you only look once!网络结构1.2. 步骤1.2.1 将图片网格化，变成S x S个cell [448 x 448 -> 7 x 7]One cell will be responsible for predicting an object as long as an object’s centerlocating in that cell.只要一个物体的中心在一个cel

目标检测之faster RCNN[2015 Ren]***回顾：***R-CNN ==> 1image/mniFast R-CNN ==> 1~2 fps1. faster RCNNBackbone(神经网络模型)pre-train(预训练)fine-tuning(微调)re-trainRPN:Region Proposal Net(区域生成网络 )代替selective searchs生成区域建议输入feature map(特征图),输出一系列ROI(感兴趣区

目标检测之RCNN1. RCNN:Regions with CNN features[2014 Ross]图片输入区域建议(region proposal)特征提取(feature extraction)分类+2次定位(classification(SVM) + detection(线性模型Linear 正则化项))2. 区域建议算法：selective search前景分...

1. 检测数据集介绍1.1. pascal VOC 2007/2012pattern analysis statical modeling and computational learningvisual object classes(模式分析静态建模与计算学习、可视化对象类别)10k张图片20~30k个对象1.2. coco大小为20G200k张图片 500~600k个对象...