Datasets:
image imagewidth (px) 896 896 | mask imagewidth (px) 896 896 | sample_id stringclasses 3
values |
|---|---|---|
00017 | ||
00017 | ||
00017 | ||
00017 | ||
00017 | ||
00017 | ||
00017 | ||
00017 | ||
00017 | ||
00017 | ||
00017 | ||
00017 | ||
0002 | ||
0002 | ||
0002 | ||
0002 | ||
0002 | ||
0002 | ||
0002 | ||
0002 | ||
0002 | ||
0002 | ||
0002 | ||
0002 | ||
0003 | ||
0003 | ||
0003 | ||
0003 | ||
0003 | ||
0003 | ||
0003 | ||
0003 | ||
0003 | ||
0003 | ||
0003 | ||
0003 |
Vivianite segmentation dataset
36 manually annotated microscopy image tiles for binary vivianite semantic
segmentation. Each tile is 896 × 916 pixels and has a corresponding binary
mask. The data covers three microscope samples: 00017, 0002, and 0003.
Structure
all/images/<tile>.png
all/masks/<tile>.png
all/metadata.jsonl
source/annotations.xml
metadata.jsonl pairs image_file_name and mask_file_name, and includes a
sample_id. All data is deliberately in one all split: tiles originating
from the same sample are correlated, so training/validation/test splits should
be made at the sample_id level rather than at the tile level.
Labels
Masks are 8-bit grayscale PNGs: black (0) is background and white (255) is
vivianite.
License
MIT. The dataset is provided as-is, without warranty. Users are responsible for validating suitability for their scientific or operational use.
- Downloads last month
- 137