RAIS

3D-UnOutDet: A Fast and Efficient Unsupervised Snow Removal Algorithm for 3D LiDAR Point Clouds

Authors: Abu Mohammed Raisuddin, Idriss Gouigah, and Eren Erdal Aksoy

3D-UnOutDet Source Code: Click Here

3D-UnOutDet Supplementary Information: Click to see Supplementary Information

Acknowledgement

Nisa B. Nar has been acknowledged for annotating some point clouds.

We thank Segments.ai for giving us free academic license to use their annotation tool for academic research. Segments.ai has been cited in the paper instead of putting their name in the Acknowledgement.

3D-UnOutDet Videos

DeSnowing with 3D-UnOutDet in WADS dataset

DeSnowing WADS with 3D-UnOutDet

DeSnowing with 3D-UnOutDet in CADC dataset

DeSnowing CADC with 3D-UnOutDet

Object Detection on De-Snowed CADC point clouds

3D-UnOutDet + PointPillers

CADC Annotations

We split the CADC dataset into three sets: training, validation and testing. Testing set has 32 sequences. One research engineer had annotated 2 testing sequences from CADC which were reviewed by one doctoral candidate. With that 2 labeled sequences, we trained 3D-OutDet to produce annotations for the remaining 30 sequences from our test set.

Click here for both Human and Machine annotations

PrePrint

Click here to see the Pre-Print