search

kaist-avelab / K-Radar

4D Radar Object Detection for Autonomous Driving in Various Weather Conditions
309 stars 44 forks source link

Dataset is too large #2

Open yangxh11 opened 1 year ago

yangxh11 commented 1 year ago

The size of K-Radar is too large. Would you consider split the lidar point cloud files separately?

DongHeePaek commented 1 year ago

Hello, @yangxh11.

I appreciate for your interest in our work.

We are uploading our dataset separately on Google Drive.

Please refer to https://drive.google.com/drive/folders/1IfKu-jKB1InBXmfacjMKQ4qTm8jiHrG_.

Thank you.

DongHeePaek commented 9 months ago

I am writing to clarify a point that the complete dataset was accessible before the camera-ready deadline. We initially shared it via a local server, acknowledging the slower download speeds from this source. To mitigate this, we offered two alternatives:

Firstly, the dataset has been made available through shipping, a method documented on our GitHub page in the "Sending 4D raw data via shipping" section. This approach has been successfully adopted by various institutions and companies.

Secondly, a partial dataset was shared via Google Drive. Due to its 2TB storage limit and the dataset's size of 16TB, including processed data like the Radar cube, full sharing on this platform was impractical. The associated costs, 119,000 KRW per year for 2TB, meant a total annual cost of 952,000 KRW for 16TB, necessitating eight members to forego personal drive usage and share costs. These constraints led us to recommend the local server or shipping method for full dataset access.

We deeply regret the inconvenience caused by the size of the K-Radar dataset and are committed to providing the best possible support for all shipping inquiries received via email. Should you require the full dataset, please do not hesitate to contact us as mentioned in the "Sending 4D raw data via shipping" section of our GitHub page.

hitbuyi commented 4 months ago

The dataset is too large, radar shoudn't have so much information, even 80% camera datasets are smaller than this, I recommand the author to reduce the dataset size