何歳に見える？ AI顔年齢診断チェッカー

What Our 68-Point Age Model Measures

Our Perceived Age Detector runs on face-api.js (TensorFlow.js with WebGL and WebAssembly acceleration). First, an SSD MobileNet v1 convolutional detector locates each face bounding box. Next, a 68-point facial landmark network maps key coordinates along the eyebrows, eye corners, nasal bridge, philtrum, lips, and jaw contour.

From the aligned and normalized 68-landmark crop, a specialized convolutional age-and-gender regression model evaluates facial proportions, geometric spacing between facial landmarks, and localized surface pixel gradients around key facial regions to calculate an estimated perceived age alongside expression probabilities. (For 3D skull-ratio classification, our Face Shape Analyzer uses Google MediaPipe's denser 478-landmark mesh.)

Tips for the Most Consistent Photo Result

• Use Even Front Lighting: Face a window or soft light source. Harsh overhead bulbs cast shadows under the eyes and nose that artificially increase estimated age by 5–8 years.
• Keep Camera at Eye Level: Hold your camera 18–24 inches away at eye height with a straight head posture to avoid wide-angle lens distortion.
• Relax Your Expression: Heavy squinting or raised eyebrows create temporary dynamic creases that the neural network reads as permanent texture lines.
• Remove Sunglasses & Heavy Filters: Thick frames occlude orbital landmarks, while beauty smoothing filters erase natural skin micro-texture.