search

jediofgever / ROS_Raw_Kitti_Player

Ros Package to access and manipulate raw KITTI data, with Camera-LIDAR sensor fusion and Perception Tasks
MIT License
71 stars 12 forks source link

Can you add a docu and code to how you obtain the maskrcnn detection images? #3

Open appinho opened 5 years ago

jediofgever commented 5 years ago

hi @appinho , Thanks for using the project and sorry for the inconvenience. it is in my agenda to add detailed documentation to obtain the maskrcnn detection images,

for the testing without installing maskrcnn and other dependencies; , I provide obtained masks for scenario; 2011_09_26_drive_0001_sync ,download the provided files here, put the maskrcnn_detections folder under 2011_09_26_drive_0001_sync/ and this should be all

but for a quick start using maskrcnn;

  1. maskrcnn that I used bases on Pytorch, and some other helper libs. A complete guide to install maskrcnn and requirements can be found here
  2. after you have maskrcnn on your system , download raw kitti images of the scenario that you would like to test. Maskrcnn provides a script to infer masks on rgb images under demo directory. I have modified that script to obtain masks in automated fashion, here is the modified script(the paths to kitti images should be changed to according to your system path)
  3. the obtained masks should be on a white background, so that when we project lidar point clouds onto this image we can differantiate objects and non-objects. to write masks on a white background ; under demo directory of maskrcnn,predictor.py should be modified as this one (very small change)

I hope this can adress the issue, let me know if you need further help