search

PRBonn / lidar-bonnetal

Semantic and Instance Segmentation of LiDAR point clouds for autonomous driving
http://semantic-kitti.org
MIT License
959 stars 206 forks source link

Kindly asking performance check TABEL 1 in RangeNet ++ research paper #54

Closed twbabyduck closed 4 years ago

twbabyduck commented 4 years ago

Dear Photogrammetry & Robotics Lab at the University of Bonn,

Based on KITTI Odometry Benchmark and the instructions inside RangeNet ++.

If iterate training dataset sequence 00 - 10 (sequence 08 for validation) to calculate std and mean, the result has different values compared to the config (img_means, img_stds) in the repository.

And after training 150 echos some of the categories might have better performance and you might want to recheck. Especially some of it has 5% difference and TABEL1 would better update a new one.

I could help with this if needed :D just let me know

huixiancheng commented 3 years ago

Could you tell me about your data of img_means and img_stds?And how about it's influcen in total mIOU?:bow: