I'm facing issues with insufficient VRAM and mismatched channel requirements.

[wbq00]

When using the latest version of the nnUNet plugin (as of January 7, 2025), I encountered two issues. First, the log indicates that the prediction process failed due to insufficient memory, causing the results array to be moved to the CPU. However, I am using an NVIDIA GeForce RTX 4060 with 8GB of VRAM, which I believe should be sufficient. Second, I received the error "RuntimeError: Given groups=1, weight of size [32, 4, 3, 3, 3], expected input[1, 1, 128, 128, 128] to have 4 channels, but got 1 channel instead." However, I have ensured that the input data is indeed 4-channel (and have also used code to forcibly convert it to confirm this). Thank you for your assistance!

[lassoan]

8GB GPU RAM just barely meets the minimum requirements. If you have are running application that uses the GPU (web browser, Zoom, Teams, etc.) then you may just run out of resources. Please exit all applications, try using a different model, use fast mode, try smaller input images, etc. and let us know if it made a difference.

If you run into any issue then all the subsequent issues may be just consequences of the previous ones. So, always focus on solving the very first reported issue, as all the others may just go away when you solve the first one.