Linketic
Could you provide your CPU memory?
When I run run_citygs_lod.sh on the residence dataset, three prune programs run parallel, and each of them uses about 170G Memory of CPU, while I only have a 500G memory, which cause an allocate memory error. @DekuLiuTesla
I check the code and confirm that viewpoint_cam = loadCam(dataset, viewpoint_cam.uid, viewpoint_cam, 1) in prune_list() causes all of the question. In loadcam(), resized_image_rgb = PILtoTorch(cam_info.image, resolution) costs about 70M and loops len(viewpoint_stack) times, which may cause memory leak and finally cost 170G memory.
Hi~, @ArSpi , thanks for your valuable feedback. Our CPU memory is over 500G, and thus didn't notice this problem. prune_list iterates over all views and it seems that images are loaded and stored in CPU memory. A possible solution is to use our loadCam_woImage in prune_list.
