// Logistics
Step-by-step instructions for preparing, testing, and submitting your final Docker container. Follow these steps in order. Test-set evaluation is manual and offline: there is no interim feedback, so a container that fails to run will not be flagged until after the deadline has closed.
(patient_id, timepoint), covering every exam in the ground-truth set:
lvef_pred, float in [0, 100]dysfunction_pred, calibrated probability in [0, 1], not a hard labelcardiotoxicity_pred, calibrated probability in [0, 1], baseline (T1) only
The scorer enforces schema strictly: missing exams, out-of-range values, duplicate
(patient_id, timepoint) keys, and missing columns are all rejected. A clean
local run is a genuine correctness signal, not just a formality.
Container harness contract — coming soon. The entrypoint invocation, input mount path, output file path, and base image/GPU requirements for the containerised evaluation harness are not yet published here. This section will be completed before the new submission deadline above. Do not finalise your Dockerfile against assumed paths; check back here first.
--network none reproduces the offline evaluation environment. If your container
depends on a runtime download (model weights, tokeniser files, API calls), it will fail
evaluation. The mount paths shown above are illustrative only; the exact paths the harness
uses will be confirmed in Step 3 once the harness specification is published.
Recommendation: to avoid any risk of exclusion, always submit as an individual using the account listed on your Team Registration Form.
Recommended naming convention (optional). Organisers recommend naming
Docker repositories and submissions using the format TeamName_TaskX_vN
(e.g. CardioVisionLab_Task1_v3). This is a recommendation for clarity only
and is not enforced.
// Q&A
Questions? echorisk.miccai@gmail.com