Abstract

This is the dataset for paper VI-BEVSEG training and testing. This dataset mostly inherits from V2X-Sim dataset. Follow the same dataset structure as the V2X-Sim dataset. Follow the V2X-Sim instruction to use. The only change we made is adding some 'h5' ground truth semantic maps. And we only keep the vehicle 1 camera information and infrastructure camera information.

Instructions:

For more details about the dataset structure, please refer to Nuscense or V2X-Sim. We only use the number 1 vehicle in the V2X-Sim dataset. Replace the folder in this dataset, and only keep the vehicle 1 six onboard cameras information, infrastructure camera information and their semantic camera information in 'sweeps' and 'samples' folder. Then replace the 'v1.0-trainval' folder with ours and put the h5 maps under V2X-Sim folder.