san9569
commented
4 months ago I think it's not a key. My problem now is that same model (same point cloud range and voxel size) trained on different training dataset shows poor results. The difference between the datasets is that one dataset is a subset of the other.
Hi,
When I use
point_cloud_range=[-54, -100, -5 54, 100, 3], the prediction results are quite reasonable.point_cloud_range=[-54, -100, -5 54, 100, 3]voxel_size=[0.075, 0.1389, 0.2]However, in the case of
point_cloud_range=[-54, 0, -5, 54, 108, 3], the results are very poor (most prediction score are close to zero).point_cloud_range=[-54, 0, -5, 54, 108, 3]voxel_size=[0.075, 0.075, 0.2]I wonder if the point cloud range should be symmetrical from zero or not.