Closed ZZy129326999 closed 1 year ago
Hello @ZZy129326999, We store one-dimensional feature vectors rather than raw points in our grids. The input points serve as training data to supervise the optimization of our map and they are not stored in any special data structure.
Hello @ZZy129326999, We store one-dimensional feature vectors rather than raw points in our grids. The input points serve as training data to supervise the optimization of our map and they are not stored in any special data structure.
Thanks for your answer. I still have two questions. The first question is whether the incremental construction of octree is reflected in the normalized point box of each frame. Once new raw points appear, a new node will be created under level-0 of octree , but if the point is not deleted in this way, will the gpu video memory keep increasing? Furthermore, will new normalized points appear, will they overwrite the original points, or will the shift point handle this problem, without explicitly updating the octree, and offset the point to simulate the incremental process? The second question is to obtain the corresponding node through the 3D coordinates of the raw points, followed by the morton code, and then the corners?
In a nutshell, when incrementally building a map, once a new point is input, its coordinates will be automatically normalized to [-1,1] to replace the old point, and will continue to increase according to the world coord?
Oh, I understand, it is the sliding window mode, thank you!
Hello, may I ask, can each leaf node store multiple points?