By logging custom data

Rerun comes with many pre-built Types that you can use out of the box. As long as your own data can be decomposed into Rerun components or can be serialized with Apache Arrow, you can log it directly without needing to recompile Rerun.

For Python we have a helper for this, called AnyValues, allowing you to easily attach custom values to any entity instance:

rr.log(
    "my_entity", rr.AnyValues(
        confidence=[1.2, 3.4, 5.6],
        description="Bla bla bla…",
    ),
)

You can also create your own component by implementing the AsComponents Python protocol or Rust trait, which means implementing the function, as_component_batches().

Remapping to a Rerun archetype

Let's start with a simple example where you have your own point cloud class that is perfectly representable as a Rerun archetype.

@dataclass
class LabeledPoints:
    points: np.ndarray
    labels: List[str])

If you implement as_component_batches() on LabeledPoints, you can pass it directly to rr.log. The simplest possible way is to use the matching Rerun archetype’s as_component_batches method.

import rerun as rr  # pip install rerun-sdk

@dataclass
class LabeledPoints:
    points: np.ndarray
    labels: List[str])

    def as_component_batches(self) -> Iterable[rr.ComponentBatch]:
        return rr.Points3D(positions=self.points,
                           labels=self.labels).as_component_batches()
…
# Somewhere deep in your code
classified = my_points_classifier(…)  # type: LabeledPoints
rr.log("points/classified", classified)

Custom archetypes and components

You can also define and log your own custom archetypes and components completely from user code, without rebuilding Rerun.

In this example we extend the Rerun Points3D archetype with custom confidence and indicator components.

#!/usr/bin/env python3
"""Shows how to implement custom archetypes and components."""
from __future__ import annotations

import argparse
from typing import Any, Iterable

import numpy as np
import numpy.typing as npt
import pyarrow as pa
import rerun as rr  # pip install rerun-sdk


class ConfidenceBatch(rr.ComponentBatchLike):
    """A batch of confidence data."""

    def __init__(self: Any, confidence: npt.ArrayLike) -> None:
        self.confidence = confidence

    def component_name(self) -> str:
        """The name of the custom component."""
        return "user.Confidence"

    def as_arrow_array(self) -> pa.Array:
        """The arrow batch representing the custom component."""
        return pa.array(self.confidence, type=pa.float32())


class CustomPoints3D(rr.AsComponents):
    """A custom archetype that extends Rerun's builtin `Points3D` archetype with a custom component."""

    def __init__(self: Any, points3d: npt.ArrayLike, confidences: npt.ArrayLike) -> None:
        self.points3d = points3d
        self.confidences = confidences

    def as_component_batches(self) -> Iterable[rr.ComponentBatchLike]:
        points3d = np.asarray(self.points3d)
        return (
            list(rr.Points3D(points3d).as_component_batches())  # The components from Points3D
            + [rr.IndicatorComponentBatch("user.CustomPoints3D")]  # Our custom indicator
            + [ConfidenceBatch(self.confidences)]  # Custom confidence data
        )


def log_custom_data() -> None:
    lin = np.linspace(-5, 5, 3)
    z, y, x = np.meshgrid(lin, lin, lin, indexing="ij")
    point_grid = np.vstack([x.flatten(), y.flatten(), z.flatten()]).T

    rr.log(
        "left/my_confident_point_cloud",
        CustomPoints3D(
            points3d=point_grid,
            confidences=[42],  # splat
        ),
    )

    rr.log(
        "right/my_polarized_point_cloud",
        CustomPoints3D(points3d=point_grid, confidences=np.arange(0, len(point_grid))),
    )


def main() -> None:
    parser = argparse.ArgumentParser(description="Logs rich data using the Rerun SDK.")
    rr.script_add_args(parser)
    args = parser.parse_args()

    rr.script_setup(args, "rerun_example_custom_data")
    log_custom_data()
    rr.script_teardown(args)


if __name__ == "__main__":
    main()