When testing the Middlebury Stereo Datasets, the calculation results were not satisfactory. The problem was that the sparse_mask.png generated by the depth map collected too many points?
hi, in paper you said you used 4-8 directions, but I can't see very clear in your code about the aggregation direction, could you explain a little about this part, about the function calcTopRowCost() and calcRowCosts()
hi, I tried out your code on a VMware virtual machine with ubuntu 16.04,
the result was good but the average execution time was about 2-3 ms/frame,
I wonder if this is normal, because on paper you said the entire pipeline can run at 5Hz
Hi,
Can you please share the script which you used to produce the gt_disparity.png image? I was trying to produce that gt_disparity.png file using my own script but it is producing very bad fusion images.
Thanks in advance.
Hi,
the masked depth used in your fusion code is from the ground truth disparity in stereo2015. For the normal sensor data, it don't have ground truth disparity image. Can you tell me how to implement the fusion algorithm on Raw data of KITTI(without ground truth disparity image, only left-right camera and lidar)?
I have transformed the 3D lidar points to depth value, and disparity = (f_u*baseline)/z as masked depth. however, the fused results are bad, worse than the original sgm. Or can you share the script which you used to produce the PMD Monstar Dataset? In you paper,you also get disparity from the depth meansurements and it seems to be a success, thanks a lot.