Part 1

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

1. Pre-processing

• 3D 사각형 영역을 Top-view형태의 8개 채널 별로 나눔 I first covert a rectangular region of lidar 3d point cloud into a multi-channel(8 channels) top view image
  • kitti dataset(2011_09_26_drive_0005_sync) 사용

2. Training Net

• I learn to use tf.py_func() to create customized tf layer. For example, in the generation of +ve (red) and -ve (gray) anchor boxes training samples for the 3d proposal net:

Part 2

• 추가 작업

  • add a “dummy” rgb image feature extraction net.

  • generate 3d proposals from top view and then project to rgb proposals again

  • add several customised op to tensorflow.

• SqueezeDet 적용

  • In order to speedup development, I am now using an existing[1-SqueezeDet] rgb kitti car car detector to replace my “dummy” rgb feature extraction net.

Part 3