Source code for blendersynth.annotations.bbox

import bpy
from .utils import project_points
import mathutils
import numpy as np
from ..blender.mesh import Mesh
from ..blender.camera import Camera
from .annotation_handler import AnnotationHandler
from typing import List, Union

BBOX_FMTS = ["x1y1x2y2", "xywh"]


def _bounding_box(
    object: Mesh,
    camera: bpy.types.Camera = None,
    scene: bpy.types.Scene = None,
    return_fmt: str = "x1y1x2y2",
    normalized: bool = False,
    invert_y: bool = True,
) -> tuple:
    """Get the bounding box of an object in camera space.
    Achieve this by projecting all vertices of the object to the image plane,
    and taking the min/max of the resulting coordinates.

    :param object: Mesh
    :param camera: bpy.types.Camera (if None, use bpy.context.scene.camera)
    :param scene: bpy.types.Scene (if None, use bpy.context.scene)
    :param return_fmt: one of ['x1y1x2y2', 'xywh']
    :param normalized: if True, return normalized coordinates (0-1) instead of pixel coordinates
    :param invert_y: if True, y is measured from the top of the image, otherwise from the bottom (Blender measures from bottom)
    """

    if camera is None:
        camera = bpy.context.scene.camera

    if scene is None:
        scene = bpy.context.scene

    verts = object._get_all_vertices("WORLD")
    coords_2d = project_points(verts, scene, camera, invert_y=invert_y)

    if normalized:
        coords_2d[:, 0] /= scene.render.resolution_x
        coords_2d[:, 1] /= scene.render.resolution_y

    xmin, ymin = coords_2d.min(axis=0)
    xmax, ymax = coords_2d.max(axis=0)

    if return_fmt == "x1y1x2y2":
        return xmin, ymin, xmax, ymax

    elif return_fmt == "xywh":
        return xmin, ymin, xmax - xmin, ymax - ymin

    else:
        raise ValueError(
            f"Invalid return_fmt: {return_fmt}. Must be one of {BBOX_FMTS}"
        )


def _bounding_boxes(
    objects: List[Mesh],
    camera: bpy.types.Camera = None,
    scene: bpy.types.Scene = None,
    return_fmt: str = "x1y1x2y2",
    normalized: bool = False,
    invert_y: bool = True,
) -> List[tuple]:
    """Run `bounding_box` for multiple objects

    :param objects: List of Mesh objects
    :param camera: bpy.types.Camera (if None, use bpy.context.scene.camera)
    :param scene: bpy.types.Scene (if None, use bpy.context.scene)
    :param return_fmt: one of ['x1y1x2y2', 'xywh']
    :param normalized: if True, return normalized coordinates (0-1) instead of pixel coordinates
    :param invert_y: if True, y is measured from the top of the image, otherwise from the bottom (Blender measures from bottom)
    """

    return [
        _bounding_box(
            obj,
            camera=camera,
            scene=scene,
            return_fmt=return_fmt,
            normalized=normalized,
            invert_y=invert_y,
        )
        for obj in objects
    ]


[docs] def bounding_boxes( objects: List[Mesh], camera: Union[Camera, List[Camera]] = None, scene: bpy.types.Scene = None, return_fmt: str = "x1y1x2y2", normalized: bool = False, invert_y: bool = True, ) -> AnnotationHandler: """ Project bounding boxes of all objects in list to all cameras provided. :param objects: List of Mesh objects :param camera: Either a single :class:`~blendersynth.blender.camera.Camera` object, or a list of them. If None, use bpy.context.scene.camera :param scene: bpy.types.Scene (if None, use bpy.context.scene) :param return_fmt: one of ['x1y1x2y2', 'xywh'] :param normalized: if True, return normalized coordinates (0-1) instead of pixel coordinates :param invert_y: if True, y is measured from the top of the image, otherwise from the bottom (Blender measures from bottom) """ return_dict = {} if camera is None: camera = bpy.context.scene.camera if not isinstance(camera, list): camera = [camera] for cam in camera: return_dict[cam.name] = _bounding_boxes( objects, camera=cam, scene=scene, return_fmt=return_fmt, normalized=normalized, invert_y=invert_y, ) return AnnotationHandler.from_annotations(return_dict, ann_type="bbox")