search

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

How to train multipleScan experiments on Semantic-KITTI #34

Closed Hao-HUST closed 3 years ago

Hao-HUST commented 3 years ago

Hi, thanks for sharing your great work!

I have been tested your train.sh to run single-scan experiments on Semantic-KITTI, but how should I run multi-scan experiments?

I notice that you add the multi-scan dataloader code on PC_dataset.py (https://github.com/xinge008/Cylinder3D/blob/6dfbbfa0ea44f1f0196951f139e756cbea5205bb/dataloader/pc_dataset.py#L130). I tried to modify the pc_dataset_type from "SemKITTI_sk" to "SemKITTI_sk_multiscan", but the train.sh failed.

If I hope to train Cyinlder3D on multi-scan LiDAR data, how should I do?

xinge008 commented 3 years ago

I just release the data processing of multi-scan point clouds; To keep the code structure clean and consistent, I do not give more modification. Because the network of multi-scan is same with single-frame, some modifications about the input are required. You can follow the error log to make it or you can wait my next update.

Hao-HUST commented 3 years ago

OK, thanks. When you plan to release the next update?

xinge008 commented 3 years ago

No specific plan now. Maybe you can watch the repo and check the notification.