news

apple's depth pro ai model sets off a revolution in ar: zero-sample learning, turning a single 2d picture into high-definition 3d in 0.3 seconds

2024-10-05

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

it house reported on october 5th that the technology media venturebeat published a blog post yesterday (october 4th), reporting that apple’s ai research team released a new ai model called depth pro, which does not require traditional camera data prediction and can achieve the desired results in a few minutes. generate detailed 3d depth maps from a single 2d image in one second.

the paper is titled "depth pro: sharp monocular metric depth in less than a second" and is a major breakthrough in the field of monocular depth estimation (monocular depth estimation). the entire process uses only one image to infer depth information.

compare the depth maps of apple's depth pro, marigold, depth anything v2 and metric3d v2. depth pro excels at capturing details such as hair and birdcage wires, producing clear, high-resolution depth maps in just 0.3 seconds, surpassing other models in accuracy and detail.

according to the paper, the model, led by the team of aleksei bochkovskii and vladlen koltun, claims that depth pro is one of the fastest and most accurate total generation systems of its type.

depth pro can generate high-resolution depth maps in 0.3 seconds on a traditional gpu, creating images with a total of 2.25 million pixels and exceptional clarity, capturing details such as hair and plants that are often missed by other methods.

what really sets depth pro apart is its ability to estimate both relative and absolute depth, a capability known as "metric depth."

depth pro does not require extensive training on domain-specific data sets to make accurate predictions. this feature is called "zero-shot learning", which means that the model can provide real measurement data, which is very useful for augmented reality (ar). applications are crucial.

depth pro is now open source on the github platform, and developers are encouraged to further explore the potential of depth pro in fields such as robotics, manufacturing, and healthcare.