Open richard5635 opened 4 years ago
See that instructions say depth is expected to be scaled by a factor of 5000 (I think you applied a factor of 1000). But I am not sure this is the problem, you should just get a smaller depth flow...
To be honest I cannot help much, I haven't run this code for ages and what I could do is the same as you can do: debug it. Add logs: see how images are once passed to the GPU, check if the solver is stopping early without performing any iteration, things like that.
Sorry.
Thanks for the feedback! Will do so. :)
Do you still have the dataset you used on this algorithm by any chance? With that, I can make sure that either my dataset's depth value is problematic or I messed the installation somewhere.
I have some of the images I used (and the results I got) here: https://www.dropbox.com/sh/lf0p6r74x1972t1/AACshZZDG-TAkrhbXC4mJ8RYa?dl=0
Let me know if you have problems to get the data.
Hi, I tested this code using MSVC 2017, and had a successful build. I tried testing this with two rgb and depth images each, obtained from KinectV2. Since their initial sizes were 512x424, I added extra lines of code to resize the images to VGA resolution (640 x 480).
RGB images are 8 bit and Depth images are 16 bit. Depth image's values are in millimeters, so they range between 1000 to 1500ish.
This is the output that I received. Basically the values of each output pixel is zero.
Is there anything else I should do with the input? Any help would be appreciated.
Thanks!