Alexander A
Alexander A
**TL;DR**: Size inconsistency is NOT an issue of the U-Net implementation for the _original_ version from the paper referenced above. The original paper used a mirror-tile strategy for input images...
> The result should be consistent after 'model.eval()' is set and same random seed is used I am not sure about this statement, the seed might be the same, but...
@ahsan-z-khan, I also have this issue. Do we have any updates on this?
What about the individual losses? Are all of them not converging? In order to generalize VoteNet to 3D rotations you would have had to adapt the bbox parametrizations. Have you...
I am playing around with KITTI right now. After some tweaks to the backbone network architecture, and some minor param changes in the Voting module, I am at around ~50...
Yes, I was planning on that when the bulk part of things I wanted to test is implemented.
Well, it is true that VoteNet does not do a fine job on KITTI dataset out-of-the-box. The problem with KITTI is not even the downsampling of point cloud (in fact,...
@lilanxiao @ilopezfr thanks guys for your feedback and all the suggestions. It has been my project for the CS230 course at Stanford. However, since I've been taking it along my...
Hi, you don't actually have to convert the coordinates to upright, as long as you are consistent with the choice of your coordinate system throughout the entire model input.
I used Octave to run the MatLab scripts to preprocess data. You will have to make one or two edits to run under Octave, though.