dataloader's workers are out of shared memory

[bjzxnsjj]

Hello,

I am very interested in trying out your placenta slice inference pipeline, but I lack a computer science background. Currently, I am attempting to run cell_inference.py on my personal PC using the sample_wsi demo data. I seem to be encountering an issue with insufficient memory, as indicated by the following error message:

RuntimeError: DataLoader worker (pid 80447) is killed by signal: Bus error. It is possible that dataloader's workers are out of shared memory. Please try to raise your shared memory limit.

I then modified the command to --nuc-num-workers 0 --cell-num-workers 0, and the output was as follows:

generating tile coordinates rows: 11, columns: 12 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 11/11 [00:00<00:00, 86399.52it/s] loading datasets datasets loaded creating dataloader dataloader ready 0%|

The main issue is that the process seems to be stuck, with the progress bar not advancing even after a long wait. Therefore, I would like to know the memory requirements for running the inference and whether GPU support is necessary.

Thank you!

[ChristofferNellaker]

Sorry for the delayed response. I am not sure what it does when the workers are set to 0, I would suggest trying with minimum 1. These are basically how many CPU threads try to help load data onto the GPU, so it is not immediately obvious to me how they run out of shared memory unless they all slam the remaining RAM you have. Try with one as this should be just one thread loading and image and passing it to the GPU.