Instructions

In essence, we only simulate noisy data cases by altering the image and LiDAR data, the ground-truth annotation will remain the same as the 3D position of the object in the surrounding worlds will not change when the sensors malfunction. And the instructions for downloading benchmark nuScenes-R and Waymo-R is listed as follows:

  1. Follow MMDetection3D to download and organize the NuScenes Dataset and Waymo Dataset.
  2. Download the corresponding noisy val pkl files below or generate it using our toolkit in Data Toolkits.
  3. Follow our Data Instructions to benchmark your models.

nuScenes-R

The total files is about 300MB.

Downloadlink for nuScenes-R (noisy val pkl file): [GitHub Release] [Goole Drive] [Baidu Yunpan]

Waymo-R

The total files is about 200MB.

Downloadlink for Waymo-R (noisy val pkl file): [GitHub Release] [Goole Drive] [Baidu Yunpan]

Note: The mask for Camera lens occlusion need to download separately. And the total files is about 10MB.

Downloadlink for the mask: [GitHub Release] [Goole Drive] [Baidu Yunpan]

Note: The links of Google Drive & Baidu YunPan is coming soon. You can use the files published in GitHub Release first.