Plate Pattern does not weigh your food through the camera.
It takes the information you provide — a description, a photo, or both — and turns it into a structured estimate of the meal's calories and macros.
The important part of that sentence is estimate.
Step 1: you provide evidence about the meal
That can be as simple as:
Chicken sandwich and a small bag of chips.
Or as specific as:
Grilled chicken sandwich on brioche with provolone and mayo. Menu says the chicken is 6 oz. Ate the whole sandwich and about half the chips.
A photo can add visual information about the foods and relative portions.
The more relevant context you provide, the fewer assumptions the estimate needs to make.
Step 2: the meal is broken into useful components
A mixed meal is easier to reason about when it is represented as components.
For example:
- chicken
- bun
- cheese
- mayonnaise
- vegetables
- chips
The system can then estimate each component rather than inventing one opaque total for "sandwich meal."
This is also what makes the reasoning inspectable.
Step 3: known information is preserved
If you provide a known quantity — "150 g chicken" — the system should use that instead of replacing it with a visual guess.
Likewise, if you give label information or a menu-listed portion, that is stronger evidence than a generic assumption.
The product philosophy is simple:
Do not infer what the user already knows.
Step 4: unknown information has to be estimated
Some details will remain uncertain:
- portion size
- cooking oil
- sauce quantity
- recipe composition
- cut of meat
- hidden ingredients
The model uses the available meal context to make reasonable assumptions.
Research on AI dietary assessment makes clear why this uncertainty exists: recognizing a food, estimating its volume and estimating its nutrients are related but distinct tasks.[1][2]
Step 5: Plate Pattern shows the result as an estimate
Estimated values are visually distinguished in the product rather than presented as if they came from a scale or label.
The Review Estimate experience can expose assumptions and component reasoning so you can understand what drove the result.
That matters because a precise-looking number can still be the result of estimated inputs.
Step 6: you can correct the record
If the system assumes full-fat yogurt and you used 0%, change it.
If the chicken portion is too small, adjust it.
If the entire meal total looks wrong and you have better information, the final record should reflect your knowledge.
Plate Pattern treats the estimate as a draft, not as authority over the person who ate the meal.
Why Plate Pattern does not publish a single "accuracy percentage"
There is no honest universal accuracy number for every kind of meal.
A packaged yogurt with a known serving is a different estimation problem from restaurant curry photographed from one angle.
Academic reviews of image-based dietary assessment show wide variation across methods, foods and evaluation conditions.[1][3]
A single marketing percentage would hide the important question: what information was available for this meal?
What improves an estimate?
The most useful additions tend to be:
- known weights
- household measures
- menu-listed sizes
- cooking method
- added oils and sauces
- hidden ingredients
- amount actually eaten
A photo is useful evidence. A description is useful evidence. Exact details are useful evidence.
Plate Pattern combines what you give it rather than requiring one input style.
What Plate Pattern is not for
Plate Pattern is a consumer meal log, not a medical nutrition instrument.
If you require clinically prescribed nutrient control, allergy management, renal-diet monitoring or another medically consequential dietary calculation, use tools and professional guidance appropriate to that purpose.
The app is designed for everyday personal nutrition awareness and macro/calorie tracking.
The design goal
The goal is not to make uncertainty disappear.
It is to make real food loggable while keeping the uncertainty visible enough that you can correct it.
That is why the workflow is:
capture → estimate → inspect → edit → save
rather than:
capture → unquestionable answer.
References
- Shonkoff ET et al. AI-based digital image dietary assessment methods compared to humans and ground truth: a systematic review. 2023. PMID: 38060823.
- Chotwanvirat P et al. Advancements in Using AI for Dietary Assessment Based on Food Images: Scoping Review. 2024. PMID: 39546777.
- Zheng J et al. Artificial Intelligence Applications to Measure Food and Nutrient Intakes: Scoping Review. 2024. PMID: 39608003.