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

What's meaning of pre-trained model #53

Closed lihao2333 closed 4 years ago

lihao2333 commented 4 years ago

If I train the pre-trained model with the semantic-KITTI, then is the final model better than pre-trained model?

tano297 commented 4 years ago

Hi, Perhaps it's a bad name for this. The purpose of a pre-trained model is either: 1) already use it with semantic-kitti, without training, or 2) use it as a good initialization to train on another dataset. I doubt you will get better results by further training, since I spent months tuning parameters for the provided models