Closed synsin0 closed 2 months ago
Hi, thanks for your attention and the good question.
For each auto-regressive prediction of the i-th frame (i={0,1,2,...}), ViDAR predicts the point clouds of (i-3, i-2, i-1, i, i+1) frame. These point cloud predictions are all in the coordinates of i-th frame.
In other words, in the main loop of the auto-regressive pipeline (code), for each iteration, the coordinates are aligned to the LiDAR coordinate of the i-th frame. For the prediction in each iteration (code), all past and future coordinates are aligned to the current lidar coordinates.
Thanks, Zetong
Thanks for your timely answer!
I'd like to ask what is the coordinate of multi-frame sensor data. I guess point clouds from past and future are transformed to the current lidar coordinate. and camera is transformed to current lidar coordinate with lidar2img. I think in your case, you do not use ego coordinate. Please point it out if I get wrong with the coordinate.