Guide
How accurate is AI calorie counting?
AI calorie counting from a photo gives an approximate number, not a measurement. Portion size, hidden oil or sugar and mixed dishes cause most of the error. We have not yet published our own test results, so this page explains what affects accuracy and exactly how we will test AI Calorie.
Get it on Google Play → Join the iPhone waitlist Available on Android. iPhone coming soon. Free to start.
What makes AI calorie estimates wrong?
The biggest causes are portion size, which a photo shows poorly, ingredients you cannot see such as oil, sugar or coconut milk, and mixed dishes where foods overlap. Lighting, camera angle and unusual dishes also matter. Because of this, a single estimate can be well off even when the foods are named correctly.
Is Cal AI accurate?
We have not tested Cal AI and have no measured accuracy figures for it, so we cannot say. Be cautious about any app that quotes a single accuracy percentage without a method. Look for the sample size, how the reference calories were measured and which dishes were tested.
What is the most accurate calorie tracker?
Weighing each ingredient on a kitchen scale and logging it from a food database is generally the most precise way to track, but it takes more time. Photo-based tracking is faster and less precise. We do not know of a trustworthy, independent ranking of apps, and we will not invent one.
How we will test AI Calorie
- Dishes: a set of common Malaysian, Indonesian and Indian meals, plus a few global dishes.
- Reference calories: weigh every ingredient on a kitchen scale and calculate calories from USDA FoodData Central values or ingredient labels.
- Photos: photograph each plate in normal lighting, from above, the way a user would.
- Runs: scan each photo three times and record each result, to measure repeatability.
- Metric: report the error for each dish and the average, with the sample size and the dishes that did worst.
- Publication: publish the method, the results table and a downloadable CSV, with the date of the test and the app version.
Results
| Dish | Reference kcal (weighed) | AI Calorie, average of 3 scans | Average error |
|---|---|---|---|
| Nasi lemak | [PLACEHOLDER: test results, to be added after the first weighed-meal test] | [PLACEHOLDER: test results, to be added after the first weighed-meal test] | [PLACEHOLDER: test results, to be added after the first weighed-meal test] |
| Roti canai | [PLACEHOLDER: test results, to be added after the first weighed-meal test] | [PLACEHOLDER: test results, to be added after the first weighed-meal test] | [PLACEHOLDER: test results, to be added after the first weighed-meal test] |
| Nasi goreng | [PLACEHOLDER: test results, to be added after the first weighed-meal test] | [PLACEHOLDER: test results, to be added after the first weighed-meal test] | [PLACEHOLDER: test results, to be added after the first weighed-meal test] |
| Chicken rice | [PLACEHOLDER: test results, to be added after the first weighed-meal test] | [PLACEHOLDER: test results, to be added after the first weighed-meal test] | [PLACEHOLDER: test results, to be added after the first weighed-meal test] |
Test date: [PLACEHOLDER: date]. App version: [PLACEHOLDER: version]. Sample size: [PLACEHOLDER: number of plates and scans].
Related: AI calorie counter · medical review · AI Calorie vs Cal AI
AI Calorie gives general nutrition information and AI-generated estimates. It is not medical advice. See medical review and our Terms of Use.