opendrivelab/openscene

3D Occupancy Prediction Benchmark in Autonomous Driving

openscene's Issues

Is there any difference between download urls of trainval sensors on Hugging Face and OpenDriveLab?

The detail of data generation pipeline

Dear OpenDriveLab Team,
Thank you for sharing your outstanding work with the community.
Will you public the details of your data generation pipeline? such as how you manage dynamic objects when merging multiple LiDAR frames in your data pipeline.
Thanks in advance.

How to generate the occ flow?

Dear all,

Thank you for building the openscene bmk for further research in autonomous driving.

Compared to other benchmarks, openscene includes occupancy flow annotation. The existing works such as FIERY, Uniad use the box annotation to compute bev annotation for one vehicle and calculate the flow between two adjacent frames. While this is a solution to create occ flow annotation, It is too coarse since the grid in one vehicle is totally the same.

I wonder if is there any difference in flow annotation in openscene benchmark and if so, could you share?

Thank you.

meaning of driving command

Hello. I'd like to ask, what do the four numbers in the driving command of the dataset represent? As I understand it, they indicate left turn, straight, and right turn. However, I'm not sure what the fourth number means.

Openscene v1.1 metadata on miniset is unavailable. Please update the google driver link.

tar: Unexpected EOF in archive

I download all data from OpenDriveLab and use the provided command line interface (CLI) for acceleration.
However, when I was decompressing the files, two of them had problems ('openscene_sensor_trainval_camera_24.tgz', 'openscene_sensor_trainval_camera_17.tgz'):
tar: Unexpected EOF in archive
tar: Error is not recoverable: exiting now