Photographs recover the food you forgot, not the portion you misjudged#
Snap every eating occasion as it happens and resolve the numbers later. The payoff, when someone measured it, was not a better portion estimate — it was a more complete record. Forty adults had their true energy expenditure measured by doubly labeled water over fifteen days while completing standard multiple-pass 24-hour recalls. Then they were shown automatic photographs of their own day and asked what they had missed. Under-reporting fell from 17% to 9% in men and from 13% to 7% in women, and the recovered energy came from 265 foods they had simply never mentioned — predominantly snacks1.
The feasibility study that preceded it shows the same thing in miniature, and more starkly. Ten adults reviewed two days of their own images after a recall interview. They added 41 foods. They adjusted 8 portion sizes2. That ratio — five foods recovered for every portion corrected — is the entire case for photo journaling and also its boundary. A photograph is close to a perfect record of whether you ate something. It is a mediocre one about how much, which is a separate skill with its own capture technique and its own ceiling.
The gap between the two numbers is the point#
It is worth sitting with why image review moves foods so much more than amounts. Forgetting a food is a binary failure of memory, and a picture repairs it completely: you either see the biscuit on the desk or you don't. Misjudging a portion is a measurement failure, and a picture repairs it only partially, because the lens recorded the top surface of the food and nothing about its depth, density, or the oil that went into the pan.
| What image review changed | Feasibility study, 10 adults | What it means |
|---|---|---|
| Foods added to the record | 41 | Memory failure — fully repairable by a photo |
| Portion sizes revised | 8 | Measurement failure — only partly repairable |
| Net change in reported energy | +12.5% (P=0.02) | Almost all of it from items, not amounts |
Data: Gemming et al., 2013; n=10 analyzed, SenseCam-assisted 24-h recall.
The reported increase averaged 1,432 kJ a day — a little over 340 calories on my conversion, not a figure the paper printed in kilocalories. That is the same order as the gap that regularly goes missing from ordinary food logs, and the reason is not coincidence: the categories that disappear are exactly the ones a photograph is best at catching, because they are eaten standing up, between meals, away from a table where any deliberate act of recording might have happened.
A photograph is the only food record you can correct after the fact. Every other kind of entry is a claim about a meal that has already been cleared away.
A photo is a capture. It is not yet a log.#
Here is where photo journaling most often goes wrong, and it is not a technical failure. It is treating the archive as the record.
When the remote food photography method was validated against food weighed on a scale, it held up well: energy intake from the photographs ran 4.7% low for dine-in meals in the laboratory, 5.5% low for take-out lunches, and 6.6% low for free-living dinners, with correlations between the photographic and weighed estimates above 0.934. Those are good numbers. But notice what produced them: trained analysts converted every image into grams and then into calories. The photograph was the input to an estimate, not the estimate.
A folder of meal pictures with no quantities attached cannot be totalled, compared week to week, or be surprising. That is the mechanism behind a pilot trial in which logging frequency predicted weight loss among people using a calorie-tracking app but not among people using a photo-based one — a result the food-diary setup guide works through in full. The lesson is not that photos are inferior. It is that the second half of the workflow is not optional: at some point in the day, every image has to become a food and an amount, even a bad one.
Two frames per meal, and the first one has a deadline#
The systematic review of image-assisted methods — 13 studies covering 10 distinct approaches in adults aged 18 to 70 — lands on one recurring failure mode. Image-based records "can be prone to underestimation" when users do not capture images of satisfactory quality before all foods are consumed3.
That is a scheduling constraint, not a photography tip. The "before" frame has a deadline measured in seconds; once the first forkful is gone, the record is of a meal minus an unknown amount. The "after" frame — the plate as you left it — has no deadline at all and does most of the work on quantity, because what you actually ate is the difference between the two. The RFPM was built around exactly this pair: a photograph of the food selection and a photograph of the plate waste.
In practice that means the discipline worth having is narrow. Get one frame before you start. Everything else — angle, light, the coin in the corner, whether you remembered to shoot the drink — can be imperfect without wrecking the record, and much of it is recoverable by typing a sentence afterwards.
The failure mode is forgetting to shoot, not shooting badly#
Martin's team ran into this before they ran into anything else. In their pilot work, "participants occasionally forgot to take photographs of their food selection and plate waste," and the fix they adopted was not better training but automated prompts — four to six alarms a day telling people to capture and send. In the main study, 118 of 120 correctly-timed prompts were answered.
That detail matters more than any of the accuracy statistics, because it identifies which step is fragile. Nobody in that trial found the camera hard to use; 93.6% of participants said they would rather photograph their food than write it down on paper. What they struggled with was noticing, in the moment, that an eating occasion was happening at all. Which is the same failure that sinks written logs, spoken logs, and every other kind — the meal that never got entered, not the one entered wrong. Voice notes have the same profile: they win on how often the meal gets captured, not on how accurately.
So build the prompt in. Photo journaling is one of the few methods where the cue can be mechanical — three or four alarms at your usual eating times will out-perform any resolution to be thorough, and unlike a written entry, an alarm you answer three minutes late still gets a usable frame.
What to do with the backlog#
By evening you have somewhere between four and a dozen images and no numbers. The resolution pass is the part people dread and it is shorter than expected, because most of a week's photographs are of food you have already resolved once.
The order that wastes least time:
- Anything with a package in the frame. These need no estimation at all; the picture is just a reminder that it happened.
- Repeats. Your usual breakfast in a photograph is your usual breakfast entry — copy it, adjust nothing.
- The pairs. Where you have a before and an after, estimate what is missing rather than what was served; a half-full bowl is a much easier judgement than a full one.
- The single frames of unfamiliar food. These are the only ones that deserve real thought, and they are also the ones where an AI estimate from the image is most useful — good at naming the food, unreliable about the amount, so type the amount if you know it.
A photo journal resolved this way is a normal food log that happened to be assembled backwards. It carries the same counting method and the same uncertainty as any other; what it does not carry is the day's worth of foods that a recall would have quietly dropped. On the measured evidence, that is worth somewhere between six and eight percentage points of under-reporting — which is a larger correction than almost anything else you can do to a food record.
FAQ#
Do I have to photograph my food before I start eating?#
The first frame does, yes. Image-based records underestimate intake when images are not captured before the food is consumed3, because a half-eaten plate hides an unknown quantity. The second frame — what is left when you finish — has no deadline and is where the amount actually comes from, since intake is the difference between the two.
Can I keep the photos and never turn them into calorie entries?#
You can, but you have an archive rather than a record. Validated photographic methods worked because analysts converted each image into grams and then calories, reaching within about 5–7% of weighed food4. A folder of images with no amounts attached cannot be totalled or compared across weeks, which is most of what a food log is for.
How many photos does one meal need?#
Two is the working standard: the food as served and the plate as you left it. Extra angles help marginally; a missing "before" frame costs a lot. If you can only manage one, take it at the start — you can describe leftovers in words afterwards, but you cannot reconstruct a portion you never saw whole.
Sources#
- Gemming L, Rush E, Maddison R, Doherty A, Gant N, Utter J, Ni Mhurchu C. Wearable cameras can reduce dietary under-reporting: doubly labelled water validation of a camera-assisted 24 h recall. Br J Nutr. 2015;113(2):284-91.
- Gemming L, Doherty A, Kelly P, Utter J, Ni Mhurchu C. Feasibility of a SenseCam-assisted 24-h recall to reduce under-reporting of energy intake. Eur J Clin Nutr. 2013;67(10):1095-9.
- Gemming L, Utter J, Ni Mhurchu C. Image-assisted dietary assessment: a systematic review of the evidence. J Acad Nutr Diet. 2015;115(1):64-77.
- Martin CK, Han H, Coulon SM, Allen HR, Champagne CM, Anton SD. A novel method to remotely measure food intake of free-living people in real-time: the Remote Food Photography Method. Br J Nutr. 2009;101(3):446-56.
