Single-stage 3D object detectors cannot project features into a common space that is adaptive for all tasks

it is impossible to project features into a common space that is adaptive for all the tasks

http://arxiv.org/abs/2608.27282v1