Missing Threshold Discrepancy with Paper

[ujaejoon]

I’ve noticed a mismatch between the value reported in the paper and what’s actually set in the repository:

• The “missing threshold” is specified as 0.2 in the Implementation Details section in the paper.
• In the KITTI config files for pedestrian, van, and cyclist, the missing threshold is set to 0.0.

When I use the official pretrained weights and change the threshold to 0.2 (as per the paper), I’m unable to reproduce the published results. Could you please clarify:

Are the config files and weights on GitHub incorrect? Or is the threshold value in the paper itself mistaken?

Thanks for your help!