My input data is point clouds, number of points in every file is different, around 5k~2w, when I set batch_size>1, seems like I need to pad them to the same length, I am puzzled about computation, does padding part cost extra computation?
Yes, you would have to pad all the point clouds to the maximum point cloud size. Apart from the very process of padding, even operations on the padded sequence will be computationally more expensive as compared to doing it without padding. The cost would be proportional to the amount of padding required for each point cloud which indirectly depends on the size of the maximum point cloud.
Thanks for your reply！