Help improve PuttLinx ball detection for everyone. This guide shows you how to set up your environment so your contributed images are as useful as possible for training the AI model.
PuttLinx uses a YOLO AI model to detect your golf ball in the webcam feed. The model was trained on a limited number of setups, so every new environment that contributes data — different mats, lighting, ball colors, camera angles — makes the detection more robust for the entire community.
More diverse training data means the model recognizes balls in more conditions and setups.
Different mats, lighting, ball types — your unique setup fills gaps the current model can't handle.
Better detection means fewer false positives, fewer missed putts, and a smoother experience for everyone.
Getting the most out of your contribution starts with how your camera is positioned and how your putting area looks. The AI model needs to clearly see a single golf ball against your mat — here's how to make that happen.
Mount your webcam directly overhead or at a steep downward angle, looking straight down at the putting surface. The ball should appear roughly circular — not oval or distorted.
Only one golf ball should be visible in the frame at any time. Remove all other balls from the camera's view before starting. This is critical for clean training data.
Ensure your putting area is evenly lit. Avoid harsh shadows across the mat, and keep lighting conditions consistent throughout the capture session.
The camera should mostly see your putting mat and the ball. Minimize clutter, equipment, feet, keyboards, and other objects in the frame.
Any putting mat or turf surface works. The more variety we get across contributors — different mat colors, textures, and brands — the better the model becomes.
During capture, move the ball naturally around different positions on the mat. Place it in corners, edges, center — give the model variety in ball placement.
Multiple balls in the frame, camera too far away showing the entire room, feet or body parts constantly in view, screen overlays or text on the camera feed, or very dark/washed-out lighting. Any of these will require us to discard the entire batch.
Here's what we're looking for — and what we can't use. Review these before starting your capture session.
Notice how each good example has the camera looking straight down, only one ball is in the frame, the putting surface fills most of the image, and the ball is clearly distinguishable from the background. The putter or club being partially visible is totally fine — that's realistic playing conditions.
Once your environment is set up, the actual capture process takes about 15 minutes. Here's the full walkthrough.
Remove all golf balls from the camera's view except the one you'll be using. Clear any unnecessary objects from the frame. Make sure your lighting is on and consistent.
In ShotBridge, navigate to the Contribute Data tab. You'll see a live camera preview — use this to verify your setup looks correct before starting. Make sure the frame shows mostly your putting mat with no extra balls visible.
Enter a name or nickname and check the consent box. This is just so we can identify contributions if we need to reach out with questions.
Hit Start Capturing and move the ball around your mat naturally for approximately 15 minutes. Place the ball in different spots — center, edges, corners, near the putter — to give the model variety in ball position and context.
When you're done, hit Stop & Prepare Upload. Review the image count, then click Upload to KW Golf. The images will be sent to us and included in the next model training run.
Only images of your putting mat and golf ball are captured and sent. No personal information, no audio, no screen content — just what the webcam sees. You can preview the camera feed before you start to verify exactly what will be captured.
