a summary for datasets in physics pixel model

Last updated on October 19, 2025

The following datasets could be in a physics pixel frame for training or inference of a physics pixel model,

in which the traing of a physics pixel model must include 1), 2), 4) and 5) of the datasets above, and may include 3) of the datasets above as optional,

and in which the inference of a physics pixel model may include only 1), 2) and 4) of the datasets above, and may include 3) of the datasets above as optional.

1) pixel dataset – visual info and other physics parameters of each pixel, including visual info (like GRBA), XY coordinates in the frame and Z (depth), near object and far object of the pixel, and other selected physics parameters (interface ID that the pixel is on, pressure, velocity and rotation of near/far object, etc);

2) object dataset – parameters of each object, including object ID and other selected physics parameters (object class, parent object, child object, velocity and rotation of object, object material, object interfaces, object mass, etc);

3) interface dataset – parameters of each interface, including interface ID and other selected physics parameters (overlap number, interface class, parent interface, child interface, etc);

4) text describing the content of the frame, the detail level of which depends on requirement or need;

5) class tree of all objects , class tree of all interfaces(optional) and class tree of all materials, and all classes include correponding physics attributes/parameters.

By the way, for object, interface and material, the model will define each one of the above 3 kinds in a frame by a unique ID and then find the closest class to match to label the each one as a class instance of the closest class.

And also, class is a comparative broader difinition to include common attributes in the class, which is a natural nature of existence or universe itself, which may help the model to understand physics nature of real world better.