VollSegZarr

Zarr-enabled volume segmentation for extremely large microscopy images

VollSegZarr is a fork of VollSeg that adds native support for Zarr format, enabling efficient processing of multi-terabyte microscopy volumes that don't fit in memory.

Features

• Zarr Format Support: Native support for chunked, compressed Zarr arrays
• Memory Efficient: Process images larger than available RAM
• 100% Backward Compatible: Works with all existing TIFF workflows
• Automatic Format Detection: Transparently handles TIFF and Zarr
• Better Compression: 60-80% smaller file sizes with Zstd
• Cloud Ready: Efficient streaming from cloud storage
• All VollSeg Modes: StarDist, UNET, Hybrid, CellPose, etc.

Installation

pip install vollsegzarr

Quick Start

from vollsegzarr import VollSeg, StarDist3D
from vollsegzarr.zarr_io import imread, imwrite

# Load model
star_model = StarDist3D.local_from_pretrained("Carcinoma_cells")

# Segment from Zarr (or TIFF - automatic detection)
image = imread("data/volume.zarr")
results = VollSeg(image, star_model=star_model, axes="ZYX", n_tiles=(4, 8, 8))

# Save to Zarr
imwrite("output/labels.zarr", results[1], compression='zstd')

See README_ZARR.md for full documentation.

Why Zarr?

For huge images like (201, 5, 7577, 7577) ≈ 114 GB:

• TIFF: 114 GB file, 2-5 min load, 114 GB RAM required
• Zarr: 20-40 GB file, < 1 sec load, on-demand memory

License

BSD-3-Clause

Credits

Original VollSeg by Varun Kapoor et al. Zarr integration for large-scale imaging.