Dear LUMIR 2025 Organizers,

I have a few questions about the challenge:

1. Ranking strategy

  • Will you publish this year’s weighting scheme? Last year’s final score was computed as 1/5 Dice + 1/5 HD95 + 2/5 TRE + 1/5 NDV.
  • The validation leaderboard now reports Dice separately for in-domain, out-of-domain and multi-modal cases—will these individual scores be used in the final ranking?
  • Will TRE and HD95 likewise be split by domain and modality?

2. Conditional mono-modal vs. multi-modal processing

The provided SynthSR + VFA baseline applies different pipelines depending on modality (registering the original image for mono-modal cases, or first converting via SynthSR for multi-modal ones). Is it permissible in the final evaluation to detect modality from the file name (e.g. LUMIR_xxxx_0000/0001.xxx) and apply separate processing streams accordingly? Or is it only allowed to use a single model and pipeline for all cases?

3. Runtime limits

What is the maximum allowed runtime per case (or overall) during the final testing phase?