qureshinomaan/Generate-Occupancy-Maps

Update-Thread

Closed this issue · 6 comments

Will Update This Thread Each once in every two days.

Sorry for the update after so much time. I am still getting used to google colab and drive. So no real work is being done right now.

Update 1 :
Was able to run PSMnet and generate a disparity image. Had to debug the code a little(indentation error, submodule not found).
But I do not understand what left and right images are. Maybe I should read the PSMnet Paper. Meanwhile @talsperre, can you provide some sample "left" and "right" image so that I can test whether I debugged the code right?

Read up on what disparity means and monocular/stereo camera settings (Will be needed for future research in RRC anyways). In this case, you can use stereo images provided in the KITTI dataset. You can find the data here: http://datasets.rrc.iiit.ac.in/KITTI/kitti_raw/ - choose any one sequence

PSMnet works fine now. Shifting task to maskRCNN right now.
Generated disparity.png

Update 2 :
Hey @talsperre, It seems that maskRCNN is working too. I used the Detectron2 provided by facebook. I will generate both the PSMnet and MaskRCNN output tommorow.
Now coming to the next part of generating the 3D Model from these, could you provide some information or should I try to find about it myself?
Thanks

You need to look at generating a 3D point cloud using disparity. Look at videos on disparity and stereo geometry for some hints. You would also need to have some knowledge about camera intrinsics and 3D transformations. Ping for further queries.