handtracking

Why we included this project

Anyone building a hand-detection feature for webcam or video feeds will get more than a tutorial here. The repo documents training an SSD hand detector with the TensorFlow Object Detection API, and the author is honest about the hardest step, choosing the right dataset: the Oxford Hands set did not work for his egocentric, table-top view, while EgoHands matched the task far better. Alongside the write-up, it includes model checkpoints you can use without retraining, plus later additions showing how to export those checkpoints for TensorFlow.js, for browser-based tracking, and to TFLite for Android and iOS. That combination makes it a practical learning path for adapting SSD to a custom object class and a reasonable starting point for a hand-interaction feature in a real app. One caveat: the code targets TensorFlow 1.4, so you may need to regenerate a frozen graph to run it on a modern setup.