Thanks for releasing this good dataset!
When I checked the flows provided by the dataset, I found that the 'terrain' voxels also have flows. the average norm of the terrain flows is 1.67 in the training set and about 0.9 in the val set. I think the terrain should be a static class, so I can't understand why they have large flow values.
I downloaded the dataset from OpenDataLab.
That's a error in the dataset. We have checked it and it only happened in the 'terrain' voxels. Please set them to 0 for a temporary solution. We will update the dataset later.
Thanks for releasing this good dataset! When I checked the flows provided by the dataset, I found that the 'terrain' voxels also have flows. the average norm of the terrain flows is 1.67 in the training set and about 0.9 in the val set. I think the terrain should be a static class, so I can't understand why they have large flow values. I downloaded the dataset from OpenDataLab.