EthanNguyen's picture
Add final reconstructed MoNuSeg splits
29073f3 verified
|
Raw History Blame Contribute Delete
1.22 kB
---
license: cc-by-nc-sa-4.0
task_categories:
- object-detection
size_categories:
- n<1K
tags:
- monuseg
- circlenet
- histopathology
- reconstruction
---
# CircleNet MoNuSeg reconstruction
Final splits: **200 training, 100 validation and 140 test images**, all 512×512 RGB.
Each ZIP includes its images and COCO-style annotations with circle centers/radii.
Extract all three together, then copy `MoNuSeg_CircleNet_Reconstructed/monuseg`
into `CircleNet/data/`.
These are new crops and case-disjoint splits, **not the recovered historical
archive or a reproduction of the paper's benchmark**. Original crop/split and
boundary conventions are unknown. Selected original cases contain 16,966
training and 6,697 test nuclei, differing from publication counts.
[Setup, reconstruction notes and scripts](https://github.com/hrlblab/CircleNet/pull/34).
Data and transformed annotations use [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).
Source: [MoNuSeg](https://monuseg.grand-challenge.org/Data/). Cite Kumar et al.,
IEEE TMI 36(7), 1550–1560 (2017), and
[the MoNuSeg challenge](https://doi.org/10.1109/TMI.2019.2947628).
CircleNet: [Nguyen et al.](https://doi.org/10.1109/TMI.2021.3122835).