Rerun

AnnotationContext

Annotation Contexts are metadata providing information to the Rerun viewer on how to interpret and display other entities. Currently, three types of annotations are supported:

  • Labels and/or colors for Rect2D and Box3D entities. These are mapped to the rectangle and box entities via their class_id components. Commonly used for object detection.
  • Labels and/or colors for segmentation images. These are mapped to the images' pixel values, which are interpreted as class_ids.
  • Labels, colors, and/or connections for Point2D and Point3D entities. These are mapped to the point entities via their class_id and keypoint_id components. Commonly used for keypoint or pose detection.

See the Annotation Context concept page for more information.

Components and APIs

Primary component: annotation_context

Python APIs: log_annotation_context

Rust API: AnnotationContext

Rectangles example 

import rerun as rr

rr.init("rerun_example_annotation_context_rects", spawn=True)

# Log an annotation context to assign a label and color to each class
rr.log_annotation_context(
    "/",
    [
        rr.ClassDescription(info=rr.AnnotationInfo(1, "red", (255, 0, 0))),
        rr.ClassDescription(info=rr.AnnotationInfo(2, "green", (0, 255, 0))),
    ],
)

# Log a batch of 2 rectangles with different `class_ids`
rr.log_rects("detections", [[-2, -2, 3, 3], [0, 0, 2, 2]], class_ids=[1, 2], rect_format=rr.RectFormat.XYWH)

# Log an extra rect to set the view bounds
rr.log_rect("bounds", [0, 0, 5, 5], rect_format=rr.RectFormat.XCYCWH)

Segmentation example 

import numpy as np
import rerun as rr

rr.init("rerun_example_annotation_context_segmentation", spawn=True)

# Create a simple segmentation image
image = np.zeros((200, 300), dtype=np.uint8)
image[50:100, 50:120] = 1
image[100:180, 130:280] = 2

# Log an annotation context to assign a label and color to each class
rr.log_annotation_context(
    "segmentation",
    [
        rr.ClassDescription(info=rr.AnnotationInfo(1, "red", (255, 0, 0))),
        rr.ClassDescription(info=rr.AnnotationInfo(2, "green", (0, 255, 0))),
    ],
)

rr.log_segmentation_image("segmentation/image", np.array(image))

Connections example 

import rerun as rr

rr.init("rerun_example_annotation_context_connections", spawn=True)

rr.log_annotation_context(
    "/",
    rr.ClassDescription(
        keypoint_annotations=[
            rr.AnnotationInfo(0, "zero", (255, 0, 0)),
            rr.AnnotationInfo(1, "one", (0, 255, 0)),
            rr.AnnotationInfo(2, "two", (0, 0, 255)),
            rr.AnnotationInfo(3, "three", (255, 255, 0)),
        ],
        keypoint_connections=[(0, 2), (1, 2), (2, 3)],
    ),
)

rr.log_points(
    "points",
    [
        (0, 0, 0),
        (50, 0, 20),
        (100, 100, 30),
        (0, 50, 40),
    ],
    keypoint_ids=[0, 1, 2, 3],
)
Privacy policy