How accurate is KalTrack?

Last updated 2026-08-26

The short version: a photo estimate is an estimate, not a measurement, and the error is not random — it usually runs low. Everything below explains why, and what KalTrack does about it.

Why a camera underestimates

A photograph records surfaces. Calories mostly hide beneath them.

• Cooking fat is invisible. A tablespoon of oil is about 120 kcal and leaves no trace in a picture once it is absorbed. The same plate of vegetables can differ by 200 kcal depending on how it was cooked, and look identical.
• Density is unknowable from above. A bowl of rice photographed from a 45-degree angle gives away its top surface, not its depth or how tightly it was packed.
• Things sit under other things. Curry over rice, dressing under leaves, butter melted into a steak.
• Portion size needs a reference. Without a known object in frame, a plate could be 22cm or 28cm, and that is a 60% difference in volume.

Independent testing has put a number on it. Research reported in August 2026 found that AI photo-based calorie-tracking tools underestimated by about 33% — roughly a third of the calories on the plate, missing (Healio, 4 August 2026). That is not a rounding error. On a 2,000 kcal day it is the difference between losing weight and not moving at all.

We are linking to research that makes our own product look worse because you are going to find out anyway, and a number you distrust for the right reason is more useful than one you trust for the wrong one. Treat any single photo estimate as a figure that is more likely too low than too high, and read the next section for what to do about it.

What KalTrack does about it

It shows you what it thinks it saw. You get the food it recognised, not just a number. If it read your bowl of dal as a cup of lentil soup, you can see that it did, which is the only way you would ever know to correct it.

Correcting a portion takes two taps. This is the intended workflow, not a fallback for when it fails. The fastest accurate log is a fast estimate you nudge, not a slow database search.

Where a figure can be read, it is read, not estimated. Scan a barcode and you get the manufacturer's own published figures. Point at a nutrition label and it reads the printed panel. Neither is a guess. Barcode lookups are free and unlimited, deliberately — we would rather you use the exact path when it exists.

It is cuisine-aware. Trackers built on American food databases quietly map unfamiliar dishes onto the nearest familiar one. That is a large, invisible error for anyone eating Indian, East Asian, or West African food. KalTrack's analysis recognises dishes as themselves.

How to get numbers you can use

1. Shoot from about 45 degrees, with the whole plate in frame. Directly overhead loses depth; side-on hides the far half.
2. Put something known in the shot when it matters. A fork, a standard mug, your hand. It gives the estimate a scale reference.
3. Correct the portion when you know better than the camera. You cooked it. You know there were three tablespoons of oil in the pan.
4. Use the barcode and label modes for anything packaged. They are exact, free, and faster than a photo.
5. Judge yourself on the trend, not the day. Consistent method beats precise method. If you log the same way every day, a 10% bias cancels out of the week-to-week comparison that actually tells you whether your weight is moving.

What we will not claim

KalTrack does not report a confidence it has not earned, and this site does not display a rating, a review count, or a user total. The app is new. When there is something real to show, we will show it.

KalTrack is not a medical device and is not a substitute for professional advice. If you have a clinical reason to need accurate intake — an eating disorder history, diabetes management, a medically supervised diet — weigh your food and work with your clinician. No app that works from a photograph is the right tool for that job, and we would rather say so here than let you find out later.