Open Lateryears opened 1 year ago
Are you working with 2D or 3D data? If you are working on 3D volumes, you will need to change the U-Net architecture to 3D
Are you working with 2D or 3D data? If you are working on 3D volumes, you will need to change the U-Net architecture to 3D
Hello, @JuliaWolleb, I appreciate your work. Regarding the consumption of GPU resources for 3D data, do you have any insights? I used 3D data as input with batch size =1 and modified the Unet model with dim=3 to accommodate the 3D architecture, but encountered out-of-memory errors when using a Tesla V100 GPU with 32GB of memory.
Hi Yes, processing 3D data might exceed the possibilities of your GPU. We solved the problem with a patch-based approach: Refer to our paper Diffusion Models for Memory-efficient Processing of 3D Medical Images, available at https://arxiv.org/abs/2303.15288
Hi Yes, processing 3D data might exceed the possibilities of your GPU. We solved the problem with a patch-based approach: Refer to our paper Diffusion Models for Memory-efficient Processing of 3D Medical Images, available at https://arxiv.org/abs/2303.15288
Hello, @JuliaWolleb . Thank you for sharing your paper PatchDDM. It seems like a promising approach for handling large datasets with limited GPU resources. May I ask if you plan to make the code publicly available anytime soon? It would be great to see your method in action and try it out on our own datasets. Thank you!
Hi JuliaWolleb: I encountered the following error while running, RuntimeError: Expected 3D (unbatched) or 4D (batched) input to conv2d, but got input of size: [1, 5, 240, 224, 139] how can I solve it , resize it after squeezing?