a multi-scan pretrained model?

xinge008 / Cylinder3D

Rank 1st in the leaderboard of SemanticKITTI semantic segmentation (both single-scan and multi-scan) (Nov. 2020) (CVPR2021 Oral)

Apache License 2.0

859 stars 180 forks source link

a multi-scan pretrained model? #112

Open caozhe1011 opened 2 years ago

caozhe1011 commented 2 years ago

I find it has 'semantic-kitti-multiscan.yaml' in your repository, do you have a multi-scan pretrained model? If I want to train multi-scan data, what should I do? Thanks.

xinge008 commented 2 years ago

I have provided the data processing of mulit-scan point cloud segmentation (for SemanticKITTI); You can check this commit https://github.com/xinge008/Cylinder3D/commit/7df59464c9b1828f9f97084997b3d47beda612cd

caozhe1011 commented 2 years ago

Thank you so much! One more question, if I want to get the IOU result of multi-scan in your paper, do I need to adjust any parameters in 'semantickitti.yaml'? The hyperparameters in 'semantickitti.yaml' are as follows:

_output_shape:

480
360
32 fea_dim: 9 out_fea_dim: 256 num_class: 20 num_input_features: 16 use_norm: True init_size: 32

max_volume_space:

50
3.1415926
2 min_volume_space:
0
-3.1415926
-4_

xinge008 commented 2 years ago

They share same hyper-parameters, except the num_class;