Petri Counter
Analysis stays local until you save feedback
Peer reviewsHow it counts

PLATE WORKSPACE

Count. Inspect. Correct.

Example · Plate 07

Preparing image…

Finding colony centers…
100%

Your example plates

Unverified counts · select to explore

Save validation feedback

Connecting to shared storage…

Choose whole plate only after checking every marker and looking for missing colonies.

Saving uploads this photo and review to the shared dataset. Check the sharing information above before saving.

SHARED VALIDATION DATA

Saved peer reviews

Each submission is kept separately, so disagreements stay visible. Saved feedback is not automatically treated as ground truth or used to retrain the counter.

Sign in to view shared reviews.

FROM PHOTO TO REVIEWED COUNT

How it counts

Image processing · Human review

A first count you can inspect, one colony at a time. The app finds visible round centers, including those inside touching groups. You check the result and make the final corrections.

An inset counting boundary leaves the dish rim outside the counting area.Boundary 01

Find the dish

The warm-colored agar helps locate an oval counting area. An inset leaves out the plastic rim and some handwriting. Choose Set dish boundary, drag inside the outline to move it, or drag a square handle to resize it. Recount to apply.

Connected regions crossing the boundary are excluded. Check the dotted outline first.

Uneven background shading is reduced to reveal dark colony shapes.PhotoColony shapes 02

Reveal the shapes

The app compensates for uneven lighting and looks for spots darker than their surroundings. Size and shape checks filter out tiny noise, likely ink, and some implausible objects.

This version is designed for dark colonies on pale agar, like the example plates.

One connected group has three rounded lobes, each marked as a separate colony center.1 connected group · 3 centers 03

Count the centers

Each rounded lobe can have its own center, even when it touches another colony. The app places one marker per accepted center and adds them up.

There is no fixed limit of two or three colonies per touching group.

WHY TOUCHING COLONIES CAN STILL COUNT

One patch. Several hilltops.

Imagine turning a colony shape into a landscape. Its edges are low; points deeper inside are higher. Each round lobe tends to form a hilltop, with a dip at the join.

The app looks for distinct peaks in that landscape. A shallow dip is harder to separate, which is where Split touching colonies helps. This is a map of distance from the edge—not the colony’s physical height.

A conceptual distance-from-edge profile has three peaks separated by two dips, suggesting three colony centers.Center 1Center 2Center 3Deeper inside the shape = higher on the map
Conceptual illustration; the app analyzes the full two-dimensional shape.

THE HUMAN PART

Inspect. Correct. Keep the evidence.

Zoom in and look for both incorrect markers and missed colonies. Use Add, Remove, and Checked to review the result. Right-click and drag pans in every tool mode.

Detected Needs a look Reviewed / added

Amber flags highlight joins, faint spots, irregular shapes, and edge detections. They are review hints, not measured probabilities of correctness.

Tune the first count

Sensitivity
Higher finds fainter spots, but may include more noise.
Split touching colonies
Higher accepts shallower joins, but may split one colony too aggressively.
Smallest colony
Filters out tiny spots that could be noise.
Exclude outer edge
Insets the counting boundary to reduce rim interference.

Recounting applies new settings and replaces manual corrections after confirmation.

Your browser does the counting

Image processing runs in the background on your device. There is no trained AI model or GPU involved. An uploaded photo stays in your browser until you choose Save feedback.

Export a CSV, annotation JSON, or marked image. Unsaved corrections stay only in this tab.

Feedback becomes useful evidence

Saving keeps the original photo, automatic detections, corrected centers, reviewer name, notes, and review scope. Separate submissions preserve disagreements.

These are candidate labels for future validation and training. Saving feedback does not automatically train or change the counter.

Know what the photo can tell you

Use an original, full-resolution photo taken directly above one dish, with even lighting and minimal glare or condensation.

Fully fused growth may hide the original colony shapes. Expert review is still needed, and real-plate accuracy has not yet been measured against verified annotations.

Back to the plate workspace ↑

built with love for our friends at Johns Hopkins