3D point cloud detection

Discussion: https://discussions.udacity.com/t/put-your-starter-code-here/234860 Dataset:http://www.semantic3d.net/

Part1:https://medium.com/@hengcherkeng/part-1-didi-udacity-challenge-2017-car-and-pedestrian-detection-using-lidar-and-rgb-fff616fc63e8

Part2:https://medium.com/@hengcherkeng/part-2-didi-udacity-challenge-2017-car-and-pedestrian-detection-using-lidar-and-rgb-bb8e28f6d987

Part3:https://medium.com/@hengcherkeng/part-3-didi-udacity-challenge-2017-car-and-pedestrian-detection-using-lidar-and-rgb-e86490774ec6

Part4a:https://medium.com/@hengcherkeng/part-4-didi-udacity-challenge-2017-car-and-pedestrian-detection-using-lidar-and-rgb-6f6a964b94b5

Part4b:https://medium.com/@hengcherkeng/part-4b-didi-udacity-challenge-2017-car-and-pedestrian-detection-using-lidar-and-rgb-9f8b910562fc

4.25-4.26: 下载了源码，尝试按照Anaconda下虚拟环境运行tensorflow的方法。具体按照 https://github.com/hengck23/didi-udacity-2017/issues/1 中zdx3578给的步骤来。但是有一些要注意到：

    1）在tensorflow虚拟环境下，opencv3.2.0 的python库名称为cv2.cpython-35m-x86_64-linux-gnu.so,所以要用cv2.so软链接一下 才能用import cv2

    2）设置了cuda的lib路径后要source一下，否则会链接不到

    3）修改了一下作者的data.py和train.py里的一系列原始路径

    4）make.sh里面的TF_INC的路径改成了tensorflow虚拟环境下的路径 ~/anaconda3/envs/tensorflow/lib/python3.5/site-packages/tensorflow/include/

     5）重新生成roi_pooling层的时候把make.sh里面的psroi_pooling的内容注释掉了

    6）运行python3 train.py的时候注意要开启tensorflow虚拟环境

4.27-4.28: 下载了Didi数据集，按照ros,velodyne，didi-visualize安装顺序安装

     1) ros:kinetic:http://wiki.ros.org/kinetic/Installation/Ubuntu

     2) velodyne: https://github.com/markstrefford/udacity-didi-competition/blob/master/velodyne-tutorials/Installing-Velodyne-Drivers-On-Ros-Kinetic-Ubuntu-16.04-LTS-Xenial.md 亲测对kinetic有效，,indigo,jade不知为何无效

     3）didi-visualize :https://github.com/omgteam/Didi-competition-solution 这个综合了好几个，好使！