Datasets:

Modalities:
Image
DOI:
License:
crisluengo's picture
Update README.md
0bac29a verified
metadata
license: cc-by-4.0

deepcell/parametric-cell-shapes Fully synthetic cell images defined by a 10-dimensional morphometry vector, for precise resolution testing.

Description: Parametric Cell Shapes (PCS) renders 256×256-pixel, 8-bit brightfield crops at 0.158 µm/px using Fourier descriptors and Perlin noise. Each image’s outline and texture are controlled by a 10-dimensional parameter vector that adjusts deviation, roughness, axis lengths, orientation, texture scale/intensity/contrast, and ring width/intensity. Optical point-spread (ring) effects and sensor noise are matched to REM-I instrument statistics. Three one-parameter grids (shape, texture, width) with 300 random samples per level enable systematic probing of embedding resolution.

The ZIP files contain the generated images in PNG format, the Parquet files contain the HFM, ImageNet and CellProfiler embeddings for those images.