6D姿态估计从0单排——看论文的小鸡篇——Learning 6D Object Pose Estimation using 3D Object Coordinates
这篇文章内容是真的多,而且大段的文字,图和公式都很少,看起来很枯燥。。。说白了就是把整个模型拆成5 5 5的125个部分,这样一个像素点都扔进去随机森林里面训练和匹配,然后查出来他最可能的类别和在一个模型中位置。就一如他的开头这句话:The key new concept is a representation in form of a dense 3D object coordinate labelling paired with a dense class labelling.(关键的新概念就是将一个稠密的3D模型坐标标记搭配稠密的类别标记组建而成的表达方式) 使用LHCF(Latent-Class Hough Forests):拆分模型成多个patch、用patch去匹配,并且使用random forest加速匹配速度,从而更快地找到对应的位置 The key new concept is a representation in form of a dense 3D object coordinate labelling paired with a dense class labelling. template-based techniques have in our view two fundamental shortcomings. Firstly, they match