query regarding downsampling data from 64 to 32

[prateep03]

Hi, I was using the convert.py/downSample utility to convert KITTI 64 resolution data to 32 resolution. If I am not mistaken, this function removes alternate(odd numbered) scanlines to convert. Is that right ?

I tried doing that and trained a model for semantic segmentation on 32 res data, converted as above. Although the model works on KITTI validation set. However, when I test on lidar data from veloview(I downloaded HDL-32 from here), the model does not generalize.

Can you please let me know if squeezeseg can work, in general, being trained on 32 res data obtained this way? Is this something you have already tried? Thanks for your help. :-)

Prateep

[BenjaminYoung29]

Hi, I was using the convert.py/downSample utility to convert KITTI 64 resolution data to 32 resolution. If I am not mistaken, this function removes alternate(odd numbered) scanlines to convert. Is that right ?

I tried doing that and trained a model for semantic segmentation on 32 res data, converted as above. Although the model works on KITTI validation set. However, when I test on lidar data from veloview(I downloaded HDL-32 from here), the model does not generalize.

Can you please let me know if squeezeseg can work, in general, being trained on 32 res data obtained this way? Is this something you have already tried? Thanks for your help. :-)

Prateep

Hi, have you solved the problem? I am new to this project. What was the result of the model on your own HDL-32? I want to try squeezeseg on my real 16 res and 32 res LiDAR. Do you have the python version code? I am not familiar with C++.