Algorithm Test Failed—Is It Due to Incorrect Input/Output Path Settings?

Hi everyone, I’m AIA_wt, and I’m encountering an issue while testing my algorithm for the PUMA Challenge (Track 1: Preliminary Test Phase). I’d appreciate your insights.

I’m using the algorithm ALALab_Test, and after submission, the result shows Failed with the error message: "The algorithm failed on one or more cases."

I suspect it might be due to incorrect input/output path settings, but even when I submitted the official baseline, I still got a False result.

My input path is set to "/test", and the output path is "/output".

Has anyone experienced something similar? Could it indeed be an issue with the input/output paths, or are there other potential reasons for the algorithm’s failure?

Any suggestions or solutions would be greatly appreciated! Thank you in advance!

Hi aia_wt,

You can access the logs of your submission here: https://grand-challenge.org/algorithms/aialab_puma_track1_1/jobs/

It seems like your code gives this error Output file 'melanoma-3-class-nuclei-segmentation.json' was not produced due to a cuda error: 2025-03-06T13:50:07.240000+00:00 RuntimeError: CUDA error: CUDA driver version is insufficient for CUDA runtime version

Maybe this helps?

Please let me know if you have any further questions!

Regards, Mark

Hi Mark,

Thank you for your reply!

The CUDA version I used is exactly the one required by the official guidelines — CUDA 11.8 with PyTorch 2.4.1. Everything works perfectly on my local environment without any errors.

I'm not sure why the submission fails with the CUDA driver version insufficient error. Could there be any specific driver version requirements on the server side that I might have missed?

Looking forward to your advice!

Best regards, AIA_wt

Hi aia_wt,

I will ask the grand challenge support whether they have an idea to what goes wrong. They normally respond quite fast.

Kind regards, Mark

Hi aia_wt,

They state that the output filepath is not correct: Are you certain you save the file to: /output/melanoma-3-class-nuclei-segmentation.json

Regards, Mark