总结一下这两天…读(shui)的paper

图片 2

  • 第二篇整体框架如上图, 将sample 的 3D的bounding box投影到2D, 然后结合class semantic segmentation, instance semantic segmentation, shape, context, location, 这些信息得到一个score,然后得到2D的proposal, 再过CNN.
  • Bounding box:(x,y,z,θ,c,t), (x,y,z) is the center of the 3D box, θ denotes the azimuth angle and c ∈ C is the object class (Cars, Pedestrians and Cyclists on KITTI) . We represent the size of the bounding box with a set of representative 3D templates t, which are learnt from the training data.
  • 这东西并不能做到实时
  • 具体没看过代码, 其实是有一些疑问的.
    3
  • 上图是第三篇的整体pipeline.
  • 就这样吧, 虽然方案不会用这些paper的, 但还是有些启发.

Leave a Comment

电子邮件地址不会被公开。 必填项已用*标注