A fully automated pipeline that generates a custom 3D-printed climbing edge from a single hand photo. Computer vision software measures your hand and a unique geometry is generated for you.

Since hand morphology varies significantly between climbers, a standard edge puts every user's fingers in a different position, often into hyperextension or causing uneven joint loading. This custom edge puts every finger in the same position.

• Consistent joint angles: comfortable, healthy joint positioning across all morphologies.
• Secure grip: the geometry keeps fingers in place without chalk while form is maintained.
• Longer sets: the comfortable geometry and the grip let you hold 30s+ sets, bringing finger training into the time-under-tension ranges used in tendon rehab and hypertrophy protocols, compared to the 6–10s a standard edge allows.
• Targeted recruitment: open-hand or crimp-specific training (FDP or FDS muscles respectively), with effective pinky engagement.
• Warmup: the fit keeps a cold-to-max ramp comfortable each session.

Computer vision is used to locate the creases of the hand. It runs in three steps:

1. Detect the credit card. It calibrates the image in mm per pixel along two axes, using PCA.
2. Detect the hand in the image and reframe the image around it.
3. On the reframed image, detect the creases of each finger.

The detector finds skin creases, but the CAD model needs joint locations in standard positions. The distance between a crease and its joint varies with the finger and the joint. I derived per-finger joint positions from published anatomical measurements. The trainer is dimensioned to those positions, not to the fold you can see.

Your photo does not have to be perfect, or taken in an exact pose. Finger splay would otherwise propagate straight into the printed part. Past a certain splay the software rotates the finger about the knuckle it estimated, back to a standard angle. You can see this in the annotated reference image (if your fingers are splayed enough).

Most of the code is checking. The measurement itself is the simple part. The rest detects problems: an occluded card, more than one hand or card in frame, bad crease detections, too much camera angle, a photo taken from too close.

Accuracy: Across a test set of 40 people, the creases are located to 0.9 mm on average, compared to my manual annotations of the images.

Repeatability: Photographing one hand eleven times, the measurements vary by 0.6 mm on average. Most of that spread comes from how the hand is posed between photos: splay of the fingers, hand being flat, etc.

Mainly recruits the Flexor Digitorum Profundus (FDP), the muscle that flexes the fingertip joint (DIP). It does so while keeping pulleys within a safe range, which makes it a good choice for rehab, prehab, and long sets. In climbing, the FDP engages whenever you're in an open-hand position, digging into a small hold, or otherwise flexing the fingertip.

Recruits the Flexor Digitorum Superficialis (FDS), the muscle that flexes the middle finger joint (PIP). This puts the fingers in a strong position, so the Crimp model holds more load. A single symmetric part works on either hand. In climbing, crimping emphasizes the FDS, particularly in a full crimp.