Checklist AICC 2026 Round 6 · Task 3
Nuclei Reconstruction
Segment cell nuclei in H&E-stained histopathology images and submit binary masks encoded as base85 PNG strings.
The task
Automating nuclei segmentation in Hematoxylin and Eosin (H&E) stained slides supports computational pathology, but staining varies across labs and scanners, nuclei overlap and touch, and some are faint or irregular.
The data are derived from the MoNuSAC (Multi-Organ Nuclei Segmentation and Classification) dataset, whose expert polygon annotations were rasterised into binary masks. Training images come with masks; test images do not. Images vary in size, organ and scanner, and the predicted mask must match each input image's dimensions.
The statement offers optional domain knowledge with sample code: colour deconvolution into Hematoxylin and Eosin channels, optical-density transformation with Macenko stain normalisation, and CLAHE contrast enhancement.
Abridged by SOTA from the official materials. The official statement has the exact rules, and it wins wherever this summary differs.
At a glance
- You get
- Training RGB JPEG images with PNG masks ({id}.jpg,
{id}_mask.png); test RGB JPEG images. - You submit
- A single CSV with columns subtaskID (always 1), datapointID (the random id from the test filename) and answer (a base85-encoded single-channel PNG mask with nuclei 255 and background 0, at the original image size).
- Scoring
- Mean Intersection over Union on the foreground class, computed per image and averaged over test images; malformed submissions score 0.
- Rules
- No pretrained models other than those provided in torchvision.
- No training on validation or test samples in any way (including unsupervised learning or computing statistics over them).
- No external datasets; no labelling of unlabelled or test data with any pretrained model (for example CLIP).
- Individual participation (maximum team size 1); at most 50 submissions per day.
- AICC contest rules (stated on each Kaggle rules page, not enforceable): no use of LLMs for writing code or getting task ideas; no internet use other than official library documentation and the contest platform; no communication with anyone during the contest; clarifications only via the #clarification-requests channel on the AICC Discord server.
- Format
- AICC Round 6, online on Kaggle, 24 Apr 2026 18:00 UTC – 26 Apr 2026 18:00 UTC.