How accurate is an AI bike fit?

What an AI bike fit measures

A camera-based bike fit films you pedalling from the side and estimates where your joints are in every frame. From those points it calculates angles: knee extension at the bottom of the stroke, knee angle at the top, hip angle, torso angle and elbow angle, plus cadence. BikeFit/LAB compares these with reference ranges for your bike type and turns the differences into saddle and handlebar suggestions in millimetres, scaled with your inseam.

How pose estimation works

BikeFit/LAB uses MediaPipe Pose, based on Google's BlazePose model, which estimates 33 body keypoints per frame on ordinary phones and computers (Bazarevsky et al., 2020). The model runs in your browser; the video is not uploaded. An angle is calculated from three points, for example hip, knee and ankle. Because each point carries a small error, angles carry an error of a few degrees too, and averaging over many frames reduces random noise but not a systematic bias.

What the research says about accuracy

We have not published a validation study of BikeFit/LAB itself, so we rely on independent research. Bini, Nascimento and Nibali (2024) compared MediaPipe with reflective markers on standard 30 fps side-view video in 26 cyclists. Typical errors were about 3–6° at the knee and hip and 5–9° at the ankle. MediaPipe showed the knee as more bent than the markers in the part of the crank cycle around the bottom of the stroke, so knee extension can be slightly underestimated. The authors concluded that knee-angle ranges used for saddle height should still be detectable, while ankle tracking was poor.

What affects the result

• Camera angle: the camera must look at the bike at a right angle. Filming slightly from the front or back shifts every angle.
• Camera height and distance: lens at saddle height, 2 to 4 metres away, with rider and bike fully in frame.
• Clothing: loose clothing hides the hip and knee; fitted cycling clothing gives more stable points.
• Light and background: even light and a calm background help the model find each joint.
• Your inseam: millimetre suggestions are scaled with the inseam you enter, so an error there carries through.

How to get a repeatable result

• Mark where the camera and the bike stand, and use the same spot every time.
• Wear the same shoes and similar clothing, and warm up for a few minutes before filming.
• Change one thing at a time and measure again, then judge the trend rather than a single number.

AI analysis or a professional bike fit?

A video analysis is a fast, low-cost way to check your position and to see whether a change moves you in the right direction. A professional bike fitter can do what software cannot: assess flexibility, injuries and foot anatomy, measure saddle pressure, look at both legs in three dimensions and adjust the bike while you ride.

Studio fit, free app or BikeFit/LAB compared →

When to see a professional

• Pain that is sharp, persistent or getting worse, numbness, or tingling: see a physician or physiotherapist.
• After an injury or surgery, or with a known leg-length difference.
• When a clear asymmetry between your left and right side keeps coming back.
• When adjustments based on measurements have not helped after a few rides.

Frequently asked questions