The data layer of BioNuclei.
BioNuclei uses microscopy datasets to study how nuclear segmentation behaves when the image domain changes. Dataset handling, split definitions and provenance are treated as part of the scientific result.
Each dataset has a defined scientific role.
Select a dataset to inspect its role, source and repository handling policy.
BBBC039
Supervised source-domain benchmark for nuclear instance segmentation and controlled source evaluation.
Open dataset pageS-BIAD634
Heterogeneous human target-domain data used for cross-domain transfer and failure analysis.
Open dataset pageS-BSST265
Target-domain dataset included in the S-BIAD634 / S-BSST265 acquisition and evaluation pathway.
Open dataset pageBBBC038
Reserved independent external benchmark for later robustness validation after the protocol is frozen.
Open dataset pageMissing images should never disappear silently.
The evaluation protocol uses explicit inventories, deterministic pairing and hard gates. A result is meaningful only when the declared dataset, split and evaluation population can be reconstructed.
Longer term validation can extend to additional external datasets once the corresponding experiments are defined and executed.
Third party data stay with their authoritative providers.
The repository stores acquisition instructions, manifests, integrity checks and processing workflows. Raw third party datasets are not redistributed.
Every result should have a traceable data lineage.
A published result should connect to a dataset version, split definition, model configuration, execution record and evaluation artifact.