This invention describes a computer method for precisely outlining different parts of an image, such as identifying objects or regions. It first processes the image through a standard neural network to extract basic features. These features are then fed into a second, "context-switchable" neural network that specifically uses depth information from small areas of the image to understand the scene better, ultimately helping the computer segment the image.
Why it matters: Filed before consumer devices commonly integrated high-quality depth sensors. The claimed use of depth information from hyperpixel districts is now significantly more practical to implement with readily available hardware and improved neural network frameworks.
AI gives you a few directions you could take this. Pick one, and we check whether your version is different enough to patent, then write the filing.
Reinvent this with AI