I have a question regarding the pyramid approach used in optical flow algorithms. Actually, the size of my memory does not allow me to store all the images in the pyramid, and I cannot average the images one by one at each level to obtain the desired sub-image. My question is, can we directly average the original image and obtain a sub-image at a specific level? If yes, how can we do it using a Gaussian filter?
The images are (grayscale) voxel-based and have a size of 2000x2000x2000. I am using a machine equipped with an Nvidia A100 GPU with 16GB of RAM. The code is written in Python.