Clarification on required output size for generated CT volumes

Hi, I'm currently working on a solution for the challenge and I would like to ask for clarification regarding the expected size of the generated volumes.

Could you please clarify:

Is there a specific required output resolution for submission?

Will the FID evaluation be performed on resized volumes (e.g., downsampled), or do we need to match the exact resolution of the real CTs?

This would help us ensure compatibility with the evaluation pipeline and optimize our generation pipeline accordingly. Thanks in advance for the clarification!

Hi Dainele,

Please see the evaluation docker on the GitHub, each metric has different preprocessing requirements that will be applied during evaluation, so generated size does not matter (although higher resolution is better for higher scores). For the FID, please again see this script: https://github.com/forithmus/VLM3D-Dockers/blob/main/ctgen_evaluation/compute_fid_2-5d_ct.py.