For the two tasks, the main difference lies in the unannotated data used for self-supervised pretraining: Task 1 provides pre-cropped patches, whereas Task 2 provides the original full 3D LSM volumes. The annotated training set, preliminary test set, and final test set are the same for both tasks.
To access the dataset, participants must first join the challenge. Afterwards, the download details will be sent to the registered email.
Task 1
Dataset for Task 1
Fixed training data conditions — ~35,050 pre-cropped patches (300×300×300 voxels) for SSL pretraining, plus 82 annotated patches for fine-tuning.
Task 2Dataset for Task 2
Open data conditions — 66 original full 3D LSM images (each exceeding 3–4×1010 voxels) for SSL pretraining. Annotated sets same as Task 1.