Hi, thank you for releasing this great work and the detailed data preparation instructions.
I am following the KITTI preprocessing commands in docs/data_preparation.md.
However, after running the filtering step for KITTI, I found that around 99% of the data is filtered out, mainly at Stage 3. This seems much higher than the comment in the documentation saying that roughly 15%–20% of the data should be filtered.
Should we adjust proj_depth_valid_num_down / proj_depth_uv_std_down thresholds, or is this usually a sign that the projected depth generation step is incorrect?
Would it be possible to provide the expected statistics or example logs for the KITTI filtering stage?
Hi, thank you for releasing this great work and the detailed data preparation instructions.
I am following the KITTI preprocessing commands in
docs/data_preparation.md.However, after running the filtering step for KITTI, I found that around 99% of the data is filtered out, mainly at Stage 3. This seems much higher than the comment in the documentation saying that roughly 15%–20% of the data should be filtered.
Should we adjust
proj_depth_valid_num_down/proj_depth_uv_std_downthresholds, or is this usually a sign that the projected depth generation step is incorrect?Would it be possible to provide the expected statistics or example logs for the KITTI filtering stage?