Get Tucked
This page lays out what we measure, how we measure it, and what we don't claim. If a figure on this page can't be explained in two sentences, it isn't in the app. That rule is the reason Get Tucked exists.
We measure the frontal area of the whole system (you, your bike, and everything strapped to it) in square centimetres. That's the size of the silhouette you present to the wind, and it's the single biggest thing you control that affects how hard you push through the air. Because we measure the whole rig, "does this bag actually cost me anything?" is a question you can answer: shoot with the fork bags, shoot without, compare.
The process is three steps. First we isolate the system (you, the bike, the bags) from the background, and throw the background away; only the silhouette survives. Then we scale it, turning pixels into real centimetres using a ruler that's already in the photo (next section). Then we project: we sum the lit silhouette into one figure, the frontal area, in cm².
Isolation is a per-pixel segmentation of the rider-plus-bike-plus-bags system against
the background. We count the foreground pixels directly from the mask, thresholding at
the mask's mid-value and striding row by row so that buffer row-padding is never
miscounted as body. Frontal area is then
foreground_pixels ÷ (mask_pixels_per_cm)², a direct pixel-to-area
conversion once the scale is known. No 3D reconstruction, no volume estimate: it is the
flat projected area the camera sees, and that is what we call it.
Power = speed × (air drag + rolling drag)
P = v × (½ × ρ × CdA × v² + Crr × m × g)
CdA (your frontal area × drag) is the one term in this equation the app measures, and the only one you change by shifting position or luggage. The rolling term is fixed by your weight and tyres; the aero term is what a smaller silhouette buys back.
FIG. 01 · Power equation. Frontal area is the one term the app measures.
A photo is just pixels until something in it has a known real-world size. Your bike is full of known sizes. The primary one is your handlebar width, a number you enter once. You tap the two bar ends in the photo, we measure that span in pixels, and that gives us pixels-per-centimetre for the whole frame. No depth sensor. No camera calibration. Just a ruler you already own, held up in the same plane as the rest of you.
Then we check our own work against a second ruler. Tap your wheel (its diameter is another dimension we can compute from your rim and tyre), and we work the scale out a second, independent way. If the two rulers agree, you can trust the number. If they disagree by more than 10%, something was mis-tapped or mis-entered, and we say so rather than hand you a confident wrong answer. (We also record your wheelbase as a reference for side-on shots.)
Taps are recorded in normalised image coordinates, converted to pixels against the
source dimensions, and the bar span is their Euclidean distance. Scale is
bar_px ÷ (bar_width_mm ÷ 10) → source pixels-per-cm, then rescaled into the
segmentation mask's pixel space by the width ratio (valid because Vision's person mask
preserves the source aspect ratio; we check that it does, to a 2% tolerance, rather
than assume it).
Triangulation, not a single point of failure. The wheel ruler is an independent scale: front-wheel overall diameter = rim bead-seat diameter + 2 × tyre width, tapped ground contact to tyre top. We compare it to the bar ruler as a relative disagreement against the bar (the bar stays authoritative, and we never swap one scale for the other without saying so), and flag anything over 10%. Wheelbase is recorded as a candidate side-on reference; today it is stored metadata, not yet an active cross-check, so we don't claim it triangulates until it does.
One known bias, stated plainly. The handlebars sit slightly forward of your torso, so the ruler is marginally closer to the camera than the body it's measuring. That makes the absolute area read a touch low. It is consistent shot to shot, so it largely cancels when you compare two positions photographed the same way, which is what the app is for.
Get Tucked does not:
Predict CdA. That is a wind-tunnel or a Notio-class sensor's job, at wind-tunnel and sensor prices. We measure area, not a drag coefficient.
Run CFD or model airflow. No simulation, no streamlines, no pressure field.
Account for yaw or crosswind. We measure your head-on projection. Real air rarely comes straight on; we don't pretend to know your day's wind.
Promise watts saved. Area is a proxy for drag, not a wattage. We will never convert a silhouette into a number of watts and hand it to you as fact.
What it is: a repeatable proxy for drag, sensitive to your position rather than the weather, and comparable shot to shot. That is a smaller claim than most of this category makes. It is also one we can defend.
No measurement is exact, and a tool that hides its own error is lying to you. So every reading in Get Tucked carries an error bar of at least ±3%, wider when a live shot's three frames disagree. When you compare two setups, we combine both readings' error bars and check whether the difference between them is bigger than that combined floor. If it isn't, we tell you the two are indistinguishable, not "1.8% better." A difference smaller than the noise is not a result; it's noise.
The habit we ask of you: re-shoot the same position before you trust a change. If the number moves when nothing about your setup did, you've just seen the floor with your own eyes.
The ±3% is a fixed, conservative minimum applied to every computed area. It stands in for the combined jitter of segmentation and scale, the two largest error sources in a photo measurement. A live shot takes three frames; when their spread is wider than 3%, the spread is the error bar, and it never goes below 3%. The floor is a model, not a per-photo measurement, and we'd rather it erred toward caution than toward flattering a difference.
For a comparison of two areas A and B, the two error bars combine in
quadrature, √(u_A² + u_B²), the standard rule for independent errors, not
simple addition. A difference |A − B| is reported as real only when it
exceeds that combined floor; for two similar areas the floor is never less than about ±4.2% of
the reading. Everything below it is shown as indistinguishable, and the
comparison screen states the floor it used, so you're never asked to take the verdict on
trust.
Where the number is weakest, so you know:
We run two automatic sanity checks and surface both. If a computed shoulder width falls outside a plausible human range (30–60 cm), that almost always means the scale was mis-tapped or the bar width is wrong, and we say so rather than report a confident wrong number. The wheel cross-check (section 03) is the second: an independent scale that flags a >10% disagreement with the bar ruler. Neither check blocks a measurement; both refuse to let a suspicious one pass unflagged.
Posture angles come from detected body landmarks in the same scaled frame: torso lean from vertical (0° upright, 90° flat), hip angle, and head drop. They inherit the same scale dependence as the area.
Honest answers to the questions people ask.
Because "measure your frontal area, informed by a defensible planar projection" doesn't fit on an App Store icon, and because we have children and this is the level of humour you get. Tucking is what you do to go faster. The name is the only joke the app tells; everything after it is straight-faced.
No. It's a tape measure, not a coach. It tells you what your position and your luggage cost in frontal area; making that number smaller is on you. What it will do is stop you from believing a change helped when it didn't.
You can measure anything you can strap to a bike, and yes, we checked. The honest, annoying answer: a bag that rides in the shadow your torso already casts adds almost nothing to your frontal area. It's hiding behind the widest part of you. So a big chest-height bar bag can come back inside the noise floor, and the app will say "indistinguishable." Read that as scientific permission. (This is a real property of projected area, not a bit we invented, which is the only reason it's funny.)
The sources behind "frontal area is a defensible proxy for aerodynamic drag": peer-reviewed aerodynamics research, plus the field tests and plain-language overviews we point people to.
If you've read this far, you already know whether it's for you.
Get Tucked · $20 · App Store →