About — ChipScan

What it is

ChipScan is a native mobile app for iPhone and Android that uses your phone's camera to count poker chips, identify their colors, and calculate a total value. It runs entirely on your device — no cloud, no account, no data collection. Line a stack up in the aim box, tap, and check the count before it joins your total.

How it works (the short technical version)

Most "AI-powered" camera apps offload work to a server. ChipScan doesn't, because it doesn't need to. The cue it relies on — the band of pale edge spots that repeats around the rim of nearly every clay or composite chip — is a geometric signal, and classical computer vision reads it on the phone itself.

The aim box is the design, not a detail. Rather than search the whole photo, ChipScan reads only what you framed. An earlier version tried to find stacks anywhere in the shot, and against a real room — a leather rail, a window, a television — it reported furniture as chips. Constraining the search to a box you drew removes that problem outright instead of trying to tune around it.

Inside the box, ChipScan:

Works out your table color from the photo itself, so there is no calibration step.
Finds where the stack meets the table, which fixes the bottom of the count.
Measures the chip's diameter, which fixes how thick one chip must be — a poker chip is about 39 mm across and 3.3 mm thick, so the two are locked together.
Counts the rim bands up the stack against that spacing, and models the top chip's face so the count stops where the stack does.
Reads the color by measuring the chip against its own pale edge spots, which cancels out the color of the room lighting.

The full technical breakdown goes into the edge-spot spiral, the two counting methods and the measured accuracy, including where it still falls down.

Then it hands you the number. The rows it counted are drawn over the photo so you can see what it did, and you confirm or correct the count before it joins your total. Counting chips from a photograph is genuinely hard — a tall stack compresses toward the top, and the spots spiral as the chips rotate — so ChipScan is built to be checked rather than trusted blindly.

Why no ML model?

Because the problem doesn't need one. Training a custom detector requires thousands of labeled images, ships a multi-megabyte model bundled with the app, eats battery, and produces opaque failures. Classical CV gets the same job done in microseconds, with every step inspectable. When ChipScan is wrong on a frame, we can dump the mask, the projection, and the peaks and see exactly why. ML failures don't give you that.

Why mobile-only?

ChipScan is iOS 16+ and Android 11+ only. No web version, no macOS app, no Linux, no Windows. The reason is simple: the app's entire reason for existing is "point a phone camera at chips." A desktop port would solve nothing — there's no desktop scenario in which you have chips, a camera, and a need to count them that a phone doesn't solve better.

Keeping the platform list short also keeps the app honest. No telemetry to "see which platform performs best." No A/B tests. No feature flags. The same code runs on every supported device, the same way, for everyone.

Privacy by design

ChipScan was built from day one around a single architectural rule: the app must not make network calls. That choice cascades through every decision:

No accounts to register → no servers to host → no databases to breach
No analytics SDK → no usage data leaving your phone
No cloud computer vision → camera frames never travel anywhere
No cloud sync → presets and scan history live on one device, period

The trade-off is real: if you switch phones, you start fresh. We think that's the right trade-off for a tool that's looking at your private game and your private chips. See the privacy policy for the longer version.

Get in touch