Severe Weather 3D Object Detection Dataset Collection

DENSE dataset

Paper address: https://arxiv.org/pdf/1902.08913v3.pdf
DatasetAddress: https://www.uni-ulm.de/en/in/driveu/projects/dense-datasets
A total of 1.4 million frames were collected at a frame rate of 10Hz.Every 100 frames are manually marked to balance scene type coverage.
The generated annotation contains 5.5k sunny frames, 1k dense fog frames, 1k light fog frames, and 4k snow/rain frames.

CADC dataset

Paper address: https://arxiv.org/abs/2001.10117
The CADC dataset is designed to facilitate research toImprove autonomous driving in severe weather conditions.This is the first public dataset to focus on real-world driving data in snowy weather conditions.
Dataset address: http://cadcd.uwaterloo.ca/
Dataset size:

Number of trips by snow level:

Data corresponding to four snowfall levels: (Intensity according to upper left, upper right, lower left, lower right)

Ithaca365 dataset

Paper address: https://arxiv.org/pdf/2208.01166.pdf
Dataset address:https://ithaca365.mae.cornell.edu/
Repeated 40 collections on the 15 km route, got severalOne hundred thousand frames of images and aligned point clouds provide amodel road segmentation label, amodel instance segmentation label, and 3D bounding box label for 7K frames. Weather conditions collected 40 times:

Dataset Experiments - Using Pretrained Point R-CNN on Kitti and Argoverse on DatasetTo test: