CaloriMateFour separate decisions sit behind a number
A useful estimate needs food identity, edible quantity, an appropriate nutrient reference and a record of what you consumed. A model might recognize chicken but confuse a fried preparation with grilled meat. It might choose the right preparation but infer the wrong weight. Those mistakes require different corrections.
When you inspect a result, read the components before the total. “Rice, chicken and sauce” with explicit portions is more informative than a single unexplained number. A total that changes when you edit a portion should remain traceable to the edited ingredient.
Give information that the photo cannot supply
For a photo, include the whole plate, a view of food height where practical, and any separate sauce or drink. Add a note when you know the preparation or weight. For text, use a description such as “180 g cooked rice, 120 g grilled chicken, sauce on the side” rather than an ambiguous dish name.
More words are not automatically more useful. The most valuable details resolve uncertainty: raw versus cooked, grams versus pieces, oil added versus oil consumed, and served versus eaten. A short correction can improve the record more than taking a brighter picture of hidden ingredients.
A worked correction, not an accuracy promise
Suppose an illustrative meal estimate contains 200 g cooked rice, while your scale measurement was 150 g. If the selected reference is 130 kcal per 100 g, that correction changes the rice component from 260 to 195 kcal: a 65-kcal difference. This is arithmetic using an assumed reference, not a measured result from an app.
Now consider an unknown sauce. Correcting the rice does not settle the sauce's recipe. Keep the known portion and uncertain ingredients distinct, instead of treating one successful edit as proof that every part is exact.
Use confidence carefully
Visual recognition confidence is not the same as nutritional certainty. A model may be very confident that it sees soup while the energy density still depends on cream, oil or dilution. Research benchmarks also have defined meals and reference methods; their scores cannot be copied onto an untested commercial workflow.
Use a range when a key input is unknown, and keep the assumptions alongside the entry. Do not turn an image estimate into an allergen check or medical decision. Package ingredients and advice from qualified professionals address questions that a meal image cannot establish.
Move from estimate to a useful diary
Check the major foods, correct the portion, add known extras and save what was actually eaten. Reuse the entry only when the recipe and serving still match. If the same lunch changes from rice to noodles, a familiar name is not a reason to keep the old numbers.
CaloriMate's store descriptions include photo/text logging, macros and meal plans. Use the current store options to inspect availability and paid features. For the underlying calculations, start with the recipe calculator and our food-science guides.
Common questions
Can a photo measure hidden cooking oil?
A photo may not reveal the amount of oil in a recipe or how much was retained. Add recipe or portion information when available and keep remaining uncertainty explicit.
What is the best correction to make first?
Correct the food identity and major edible portions, then known energy-dense extras such as oil, dressing or sauces.
Sources & methods
- Nutrition5k: original CVPR research paperControlled benchmark separates portion information from nutritional composition; its results do not measure CaloriMate.
- USDA FoodData Central data documentationDifferent food data types have different provenance, preparation detail and completeness.
- CaloriMate: Google PlayPublisher-described photo/text logging, nutrition tracking, meal plans and water tracking.
- FDA: food allergiesIngredient disclosure, allergens and cross-contact.