Open xuzukang opened 8 months ago
Detection tasks usually evaluate boxes in lidar coordinate system, and prediction tasks usually evaluate trajectories in ego coordinate system. We guess that making the prediction head directly output coordinates in the ego coordinate system doesn't hurt the model performance.
Hello, I noticed in the code that after the memory bank gets the query, the coordinates output by the connection detection head are in the lidar coordinate system. And the coordinates output by the connected prediction head are in the ego coordinate system? Why is there such a situation where the coordinates are not uniform?