Way to generate tight 2D bounding boxes for autonomous driving labeling

A non-transitory machine-readable medium having instructions stored therein, which when executed by a processor, cause the processor to perform operations for perceiving obstacles of autonomous driving, the operations comprising:

• generating a two-dimensional (2D) segmentation image of a three-dimensional (3D) scene comprising one or more objects by rendering the 3D scene with a segmentation camera, wherein each of the one or more objects is rendered in a single respective different color;
• identifying one or more visible objects in the 3D scene among the one or more objects based on the segmentation image, further comprising:
• determining a quantity of visible pixels for each of the objects in the 3D scene, wherein all pixels present in the segmentation image associated with an object are visible pixels, and wherein a correspondence between a pixel and its associated object is identified based on a color of the pixel and a single color of the associated object;
• determining, for each of the objects, whether the quantity of visible pixels associated with the object is greater than a predetermined threshold;
• and
• identifying an object in the 3D scene as a visible object in response to determining that the quantity of visible pixels associated with the object is greater than the predetermined threshold;
• generating a 2D amodal segmentation image for each of the visible objects in the 3D scene separately, wherein each amodal segmentation image comprises only a single visible object for which it is generated;
• and
• generating a 2D bounding box for each of the visible objects in the 3D scene based on the amodal segmentation image for the visible object.

A data processing system, comprising:

• a processor;
• and
• a memory coupled to the processor to store instructions, which when executed by the processor, cause the processor to perform operations for perceiving obstacles of autonomous driving, the operations including
• generating a two-dimensional (2D) segmentation image of a three-dimensional (3D) scene comprising one or more objects by rendering the 3D scene with a segmentation camera, wherein each of the one or more objects is rendered in a single respective different color,
• identifying one or more visible objects in the 3D scene among the one or more objects based on the segmentation image, further comprising:
• determining a quantity of visible pixels for each of the objects in the 3D scene, wherein all pixels present in the segmentation image associated with an object are visible pixels, and wherein a correspondence between a pixel and its associated object is identified based on a color of the pixel and a single color of the associated object;
• determining, for each of the objects, whether the quantity of visible pixels associated with the object is greater than a predetermined threshold;
• and
• identifying an object in the 3D scene as a visible object in response to determining that the quantity of visible pixels associated with the object is greater than the predetermined threshold,
• generating a 2D amodal segmentation image for each of the visible objects in the 3D scene separately, wherein each amodal segmentation image comprises only a single visible object for which it is generated, and
• generating a 2D bounding box for each of the visible objects in the 3D scene based on the amodal segmentation image for the visible object.