Become an expert in your own dataset

Image datasets are rarely as clean or consistent as they appear. PixelPatrol scans your collection - from a single image to millions of files - and produces a shareable table you can explore as an interactive report: file metadata, pixel statistics, quality metrics, and per-dimension breakdowns. Get immediate results, compare conditions, catch outliers, verify batch consistency, and get the full picture before you use your dataset.

View Example Report

Your data stays local - nothing is uploaded.

From a PixelPatrol Report

Some Example Visualizations

How it works

Our automated pipeline

Set up a directory and a few parameters. PixelPatrol handles the rest.

Pipeline: Pick images → Process images → Collect metadata and metrics → Visualize as interactive report

Point to any folder. Common formats supported, subdirectories walked automatically.

Results written to a compact, portable Parquet file - ready to share or load into other tools.

Files scanned in parallel. Pixel statistics, quality metrics, and metadata extracted in one pass.

Open in any browser. Group, filter, compare distributions, inspect images.

Key Features

Built for real-world datasets

Shareable reports

PixelPatrol creates a parquet file that anyone can open in the hosted viewer - no installation required.

Group comparison and filtering

Compare conditions with violin plots, histograms, and distributions. Change groupings interactively - no reprocessing needed.

Per-dimension statistics

Statistics per Z, T, C slice - not just per file. Detect drift, photobleaching, and channel artifacts at a glance.

Parallel processing and HPC support

Runs across all CPUs using Dask. Connect to an HPC cluster to process terabyte-scale datasets in minutes.

Any dataset size and format

Large images processed in chunks. Container files (LMDB, OME-TIFF) supported, each sub-image as an individual entry.

Extensible by design

Add loaders, processors, and viewer widgets as Python packages. Discovered automatically at runtime.

Performance

Distributed processing at scale

Benchmarks run on HPC infrastructure across multiple nodes.

Many small images

1.1M files · 7.8 GB

~50-400 px per X/Y dimension
Wall time: 32m 44s · 558 files/s
20 workers · Peak worker RAM 0.73 GB

Large 3D volumes

3 files · 1,200 GB

9,000 × 8,000 × 2,000 px per image
Wall time: 19m 44s · 129.6 tasks/min
100 workers · Peak worker RAM 1.9 GB

Large 2D multichannel

2 files · 596 GB

50,000 × 50,000 × 40 px per image
Wall time: 6m 49s · 77.8 tasks/min
100 workers · Peak worker RAM 0.39 GB