Logging meals from a photo: how accurate is it, really?

Snapping a photo of your plate and getting an energy estimate back feels almost too easy. So it's fair to ask how much a photo can really tell an app about your meal. The honest answer: quite a lot about what's on the plate, less about how much, and very little about what you can't see. That's still useful, as long as the app is upfront about the uncertainty.

What a photo can and can't see

Researchers have worked on image-based dietary assessment for well over a decade. A 2024 scoping review in the Journal of Medical Internet Research traced how these systems moved from hand-crafted algorithms to deep learning, with big gains in recognising foods. The same review is clear about where the hard problems remain.

What photos do well:

• Identifying common foods and dishes.
• Spotting the main components of a plate: a grain, a protein, vegetables.
• Telling a side salad from a main-course salad.

What photos struggle with:

• Portion size. A single flat photo distorts size, because camera distance and angle change how big food looks. The review calls portion estimation a challenging task.
• Hidden ingredients. Oil, butter, sugar and sauces are often invisible but can change the energy of a dish a lot.
• Density. A cup of rice and a cup of salad leaves look similar in volume and are very different in energy.

How accurate is it, really?

Studies comparing photo-based apps with weighed food records, the research gold standard, tend to land in the same place. A 2026 study in Nutrients compared an image-recognition food app with weighed records and found moderate agreement, with energy and major nutrients systematically overestimated. The authors concluded the app was useful for watching overall dietary trends over time, but called for caution where precise individual numbers are needed. It was a small pilot study, so the exact figures won't apply to every app.

That's a reasonable summary of the whole field. Photo estimates are good at telling you roughly where you are and which way things are moving. They're not a laboratory scale.

It helps to compare photos with the alternative, though. Logging from memory has its own large errors. In a classic 1992 study in the New England Journal of Medicine, a small group of participants under-reported what they ate by about 47% on average when their self-reports were checked against measured intake. Recalling food accurately is hard for everyone. A photo taken at the table removes the memory part of the problem.

Why ranges beat false precision

An app that says your lunch was 642 kcal is implying an accuracy no photo can deliver. An app that says "about 600 to 750 kcal, medium confidence" is telling you the truth.

Ranges matter for practical reasons:

• They show you where to look. A wide range on a pasta dish tells you the sauce and portion are the uncertain parts. Adding "cooked with a tablespoon of olive oil" narrows it.
• They stop you chasing noise. If two similar dinners come back as 640 and 700, a range shows they're essentially the same meal.
• They keep the focus on patterns. Your typical breakfast, how often you eat before training and whether you're eating enough on long days all matter more than any one number.

The same principle applies to how an app talks about food generally. Food isn't good or bad, and an estimate isn't a verdict. It's information you can use or ignore.

How to log a meal in under ten seconds

The best log is the one you keep. A few habits make it quick:

• Photograph before you eat. One photo from slightly above, with the whole plate in view. A fork or your hand in shot gives a sense of scale.
• Use a barcode for packaged food. The label is more accurate than any photo, so scan when you can.
• Save favourites. Most people eat a small set of breakfasts and lunches. Saving them turns a repeat meal into one tap.
• Describe what the camera can't see. A few words like "with butter", "large bowl" or "shared with a friend" make the estimate far better.
• Log the occasion even if it's rough. A rough log of every meal tells you more about your pattern than a perfect log of half of them.

Using the log for training

For training, the useful questions are simple. Are you eating enough on days with long or hard sessions? Is there protein in most meals? Do you eat something before morning sessions, or train on empty and feel flat? A photo log answers those questions well.

If you're preparing for a mixed event like Hyrox, for example, fuelling around running and lifting days makes a real difference. Hyrox for beginners covers the basics. And a good coach uses food data the same way it uses sleep: as context for the plan, which is how an adaptive plan should work.

How Stwong logs meals

In Stwong, you can log a meal from a photo, a barcode, a favourite or a short description. Estimates are shown as ranges with a confidence, never a single falsely precise number. Estimation runs on your iPhone first and only uses the cloud with your consent. Photos are processed in memory and never stored.

You can also choose how your coach talks to you about food and training: gentle, straight or blunt, with short or fuller explanations. Picking a coaching style goes into why that matters.

Sam, the everyday coach, keeps it warm and plain and never comments on your body.

Three quiet days in the log, and that's fine. A photo of dinner is enough. Small and steady.

If you want meal logging that's quick and honest about what it knows, find out when Stwong arrives.