I am trying to reimplement the behaviour of ROS rviz markers in rerun. I know there's an example out there, but this question is about a specific behaviour of said markers. https://docs.ros.org/en/noetic/api/visualization_msgs/html/msg/Marker.html Notice that the message has

bool frame_locked                    # If this marker should be frame-locked, i.e. retransformed into its frame every timestep

If this is set to true, then when the marker is published like an entity on the entity tree in rerun, it'll follow Transform3D and get propagated. The trouble is, is there support for the behaviour when the field is false? (I.e. the marker should "freeze" in place relative to the root coordinate frame, regardless of where the parent is moving. If there is no such capability, what's the best way to, at runtime, do a lookup of the transform tree to figure out the transform from root to leaf entity so that the entity can be placed relative to the root instead, to freeze it?

Hi, when 3D entities are projected to camera image, e.g. Boxes3D, is camera distortion considered? If I wanna implement the projection/unprojection, which code should I modify?

Hi all, Is there a way to persistently display the AnnotationContext legend in the viewer UI, either through a panel setting or via blueprints? Currently (video): I can see the annotation colors/labels by clicking on an entity in the hierarchy and expanding the "context" section in the Selection Panel. However, the legend disappears as soon as I click elsewhere (e.g., selecting objects in the 3D view). I'm looking for a way to keep the annotation legend always visible while I interact with other entities in the scene (even if I toggle off labels on top of each object). I couldn't find a relevant panel or blueprint option for this. Has anyone found a workaround? Thanks! https://cdn.discordapp.com/attachments/1441459045197283420/1441459046568689664/Screen_Recording_2025-11-21_at_11.01.28_AM.mov?ex=6921ded9&is=69208d59&hm=6a2cf24230704a2eba05d93a647937789ea5066963ff4817b24cf07edf24be5c&

I recently upgraded my rerun-sdk from 0.22.0 to 0.26.0. Prior to this change, when projecting 3D bounding boxes into my camera frames, they aligned well. After this change, they are not even close, they don't even land in the image plane. The 3D view still shows that all my camera frames are properly aligned w.r.t the 3D bboxes. Is there anything that changed between these two versions that might explaint his? Here are some code snippets showing my implementation:

def __init__(self):
    rr.log(WORLD_FRAME, rr.ViewCoordinates.RIGHT_HAND_Z_UP, static=True)

def _log_ego_pose(self):
    rr.log(
        f"{WORLD_FRAME}/{EGO_VEHICLE_FRAME}",
        rr.Transform3D(
            translation=ego_from_world_matrix[:3, 3],
            mat3x3=ego_from_world_matrix[:3, :3],
        ),
    )

 def _log_image_data(self):
      for i, (img, K_mat, c2e) in enumerate(zip(images, intrinsics, camera2ego)):
          cam_entity = f"{WORLD_FRAME}/{EGO_VEHICLE_FRAME}/camera_{camera_names[i]}"
          rr.log(cam_entity, rr.Transform3D(translation=c2e[:3, 3], mat3x3=c2e[:3, :3]))
          rr.log(
              cam_entity,
              rr.Pinhole(image_from_camera=K_mat[:3, :3], resolution=[img.shape[2], img.shape[1]]),
          )
          rr.log(f"{cam_entity}/image", rr.Image(image_np).compress(jpeg_quality=JPEG_QUALITY))

I'm looking at https://rerun.io/docs/howto/logging/send-columns#updating-an-image-over-time-in-a-single-operation which is pretty handy for seeing how to use the Columnar API to send images. However, with the "regular" API I am able to send a compressed image by adding

.compress()

-- is there an equivalent here when using

send_columns()

? Thanks!

Hello all! I'm just getting started with rerun and I've been playing with the URDF file loading support from 0.24 onwards. Have some questions/feedback. Based on the [animated URDF example](https://github.com/rerun-io/rerun/blob/7ac9d608c483c5a5c58e883ad9a0cecc3881f82c/examples/rust/animated_urdf/src/main.rs), am I correct in assuming the best path forward for updating the position of a model with joint angles over time is: For each link, * Compute the forward kinematics (pose of the link) * Find its hierarchical path in the log * Set the new transform Some additional things I ran into: 1. Our URDFs definitely have collada (

.dae

) files and I see those aren't yet supported, so +1ing that existing issue 2. We are doing some processing of the file (via

xacro

) so we actually already have the string contents of the file. Is there a way to avoid having to write these contents to a temp file and then load them in rerun via

log_file_from_path?

It doesn't seem like

log_file_from_contents

did what I expected, but maybe I'm using it wrong? 3. In the animated URDF example, there is the

UrdfTree

utility which allows traversing the URDF and getting the correct paths for setting transforms. But there are no Python bindings to this at the moment, from what I see -- is that right? I can work around this by using other libraries like

yourdfpy

or

pinocchio

on the Python side, but of course they don't offer the same

get_joint_child()

and

get_link_path()

functions right out of the box to get the hierarchical structure that rerun puts the entities in. 4. +1 to another issue I found to configure whether to show visual and/or collision meshes more easily. Thanks!

Hi, I try to set Boxes3D.colors with alpha value, but is seems not working. Any idea? https://cdn.discordapp.com/attachments/1436363152890658949/1436364282639028276/Screenshot_2025-11-07_at_15.38.14.png?ex=690f55fb&is=690e047b&hm=90489447ee4ee12716ac31cafbcd1ed7e0d33b890ca5be0da8223b2e56c781b1&

Hi all, A similar issue (https://github.com/rerun-io/rerun/issues/6600) in 2D view has been reported and has been resolved. But I'm unsure how to resolve flickering in 3D view, and I couldn't find any sample code of using

DrawOrder

. The figure shows a occupancy grid in 3D view, where black boxes (obstacles) overlapped with traversable grids. https://cdn.discordapp.com/attachments/1434794293272248520/1434794526848979084/Screenshot_2025-11-03_at_14.29.34.png?ex=6909a008&is=69084e88&hm=cc822a2bbef85bd117337223a0e1e5cc96d5c23f4f59042aea811fe2d5547c5e&

Hi all, I have tried install

rerun-sdk

via

conda

and

uv

(a package manager), and the later failed. Specifically: 1. Conda (success)

bash
➜ which python
/opt/homebrew/Caskroom/miniforge/base/bin/python
➜ which pip
/opt/homebrew/Caskroom/miniforge/base/bin/pip
➜ pip install rerun-sdk
...
➜ python -c "import rerun.blueprint"
# OK

2. UV (failed)

bash
➜ uva
[INFO] Activate Python venv: .venv (via .venv/bin/activate)
➜ which python
/private/tmp/rr-test/.venv/bin/python
➜ which pip
/private/tmp/rr-test/.venv/bin/pip
➜ uv add rerun-sdk
Resolved 8 packages in 361ms
Installed 6 packages in 25ms
 + attrs==25.4.0
 + numpy==2.2.6
 + pillow==12.0.0
 + pyarrow==22.0.0
 + rerun-sdk==0.26.1
 + typing-extensions==4.15.0
➜ uv add rerun
Resolved 9 packages in 285ms
Installed 1 package in 4ms
 + rerun==1.0.30
➜ python -c "import rerun" # rerun can be imported
➜ python -c "import rerun.blueprint" # rerun.blueprint failed
Traceback (most recent call last):
  File "<string>", line 1, in <module>
ModuleNotFoundError: No module named 'rerun.blueprint'

For both methods, the`rerun` cli and python library are successfully installed, but

rerun.blueprint

failed. I'm wondering is there any limitation if

rerun

is installed via a python package manager? Thanks.

Hi all — hopefully a quick (maybe dumb 😅) question: Is there a way to toggle visibility in the Python API for a Horizontal or Vertical rerun.blueprint panel? I tried something like this:

python
rrb.Horizontal(
    rrb.TimeSeriesView(origin="/error1", visible=True), # <- this is ok
    rrb.TimeSeriesView(origin="/error2", visible=False),
    rrb.TimeSeriesView(origin="/error3", visible=True),
    column_shares=[1, 1, 1],
    visible=False,  # <- this doesn't seem to be accepted
    name="Metrics",
)

but got this error:

python
Horizontal.__init__() got an unexpected keyword argument 'visible'

I might have missed it in the docs — I noticed there’s an eye icon next to Horizontal and Vertical containers in the GUI, similar to individual content containers, that lets me toggle visibility, so I was wondering if there’s a programmatic equivalent. Rerun version: 0.26.0 Python version: 3.13.7 Linux Thanks!

I can't update to the latest version with my conda workflow. Do you plane to make it available ?

Hello! I'm using rerun for a ROS2 project but the build times are very long (10min+!) Any tips?

On the topic of set_sinks() with GrpcSink(), I'm seeing that my application memory continues to grow unbounded if there is no server for the sink to connect to...does anyone know if there is a way to limit the amount of data that GrpcSink() can buffer when it is failing to connect to a server?

I have been studying the Rerun Rust SDK examples to find a way to retrieve the latest during replay. So far, I have studied the ViewerEvents and ChunkStoreEvents structures. I've observed that the ViewerEventKind::TimeUpdate successfully provides me with the current time as the replay progresses, but it does not provide the actual data (e.g., an rr.Image or sensor data) for that timestamp. I can do a viewer extension as in viewer_callbacks example and in eframe::App::update query the latest instance for entity for the time received from ViewerEvents. But this seems quite clumsy. I do not really want to extend the viewer. What I really need is an event that comes every time a specific entity instance updates during replay. Is there such an event? Could someone please point me in the correct direction or confirm if this type of event is not directly supported?

Hi, I've been using Rerun to generate visuals for my AV ML training runs. We use ML Flow to log all our results, so I'm also using ML Flow to log rrd files for each training and validation step. Currently, to view the generated rrd files, I need to download them to my computer from the ML FLow UI and then launch Rerun and select that file. I know we're able to directly view rerun files from the browser, so I'm wondering if there is a way to log artifacts in ML Flow in a way that I can just click on my rrd file in ML FLow and it open up a viewer in the browser? Thanks in advance for the help!

I can make a reproducible example on github unless this is a known problem already, and can provide a video immediately. I'm using rust in Linux as the client, 0.25.1 rerun-sdk from pip install. I think it may also involve multiple processes logging at the same time, then recycling one or the other. The initial few connections work smoothly, then reconnecting one and the frame rate in rerun goes super low, and typically I have to kill everything and restart rerun to get an acceptable frame rate and to be able to use the rerun gui.

Hi There 👋 When I load a recording with

rr.dataframe.load_recording

,

recording.schema()

doesn't list the columns related to VideoStream. Is it possible to extract a video from an .rrd file? Thank you!

Hi everyone! I'm currently using Rerun at work for robotics, and my team could really benefit from the ability to visualize our protobuf messages with a tree view, so I went ahead and filed a feature request here: https://github.com/rerun-io/rerun/issues/11413. When browsing recent discussions it seems like there may already be pull for this (I saw others asking about custom protobuf schemas) so I wanted to know if anyone is already tracking a similar feature and when it would likely make it into an upcoming release? If it's not on anyone's immediate radar then we may try to contribute it ourselves. If that's the case then would appreciate any pointers on what needs to be done to extend the data tree view to user-logged structs.

Hi! I am trying to use the experimental version of rerun for mcap visualization, and when I ran it on my mcap file, I got this error : (venv) allenbiju@Allens-MacBook-Pro-2 viz % rerun cooperscenes_test_full_revised.mcap [2025-09-27T00:56:12Z INFO re_data_loader::load_file] Loading "cooperscenes_test_full_revised.mcap"… Error: "cooperscenes_test_full_revised.mcap" -> No data-loader support for "cooperscenes_test_full_revised.mcap" My mcap data is protobuf data, is that the reason I can't run it? Or is there a different reason? Also if it is because my data is not in the correct format, what is the correct format? Thank you!

Hi, Rerun noob here! I'm running into an issue while tryning to stream geotagged images using rr.Image. The fram transitions are really glitchy, and there's a noticeable delay between images in the attached video. This happens even after i downsized my images from 2448x2028 px down to 256x256. Setup: - Rerun: version 0.24.1 - Python: 3.10.17 - OS: Ubuntu 24.04.3 LTS - Kernel: 6.8.0-83-generic Since i'm still learning Rerun, it's totally possible I'm missing something obvious. Would love some guidance on whether this is a known issue or if I'm doing something wrong https://cdn.discordapp.com/attachments/1421137493885059093/1421137494539501610/simplescreenrecorder-2025-09-25_14.09.08.mkv?ex=68d7f0ee&is=68d69f6e&hm=c5e2ecc642997b1b523d1c126f32f002978623006b8c17f1eadfe2c5b942a728& https://cdn.discordapp.com/attachments/1421137493885059093/1421137494883172443/message.txt?ex=68d7f0ee&is=68d69f6e&hm=fde7d9777b86aa01e46f71bcd8eb13151871fb2304bdbc29ea71165677aa89c8&

Heya! I am looking to visualize some trace data for processes that are running on my robot. Do we have any suitable archetypes or visualisers that might be able to show something like this hierarchical trace (screenshot added)? Like a plot type or something? https://cdn.discordapp.com/attachments/1420551819376988190/1420551820090282095/trace-example.png?ex=68d5cf7a&is=68d47dfa&hm=a83eece918a1dd319d634df668f080c1bc59d6a148d45970a610c93276380bf9&

If you load a urdf whose mesh objects have a scale different than 1,1,1, the scaling is not done correctly and the meshes are displayed in their original size. To Reproduce: log a urdf normally log a urdf after chaging the scale parameter

from urdf_parser_py.urdf import URDF
robot = URDF.from_xml_file(self.__urdf_path)
for link in robot.links:
    if link.visual is not None:
        geom = link.visual.geometry
            if geom.scale is not None:
                geom.scale = [2.0, 2.0, 2.0]
rr.log_file_from_contents(
           urdf_path,
           robot.to_xml_string().encode("utf-8"),
           static=True,
)

Expected behavior The meshes have different sizes depending on the scale OS: Ubuntu 22.04 Rerun version: 0.25.1 Additional context Many STLs are in mm so they need scale=0.001 0.001 0.001 to be properly displayed as URDFs

Moving to trunk’s discord server.

Is there some easy way to invert a colormap? Most recent depth estimation methods use red for near and blue for far but the

"Turbo"

colormap does it the other way around. In Matplotlib I could use

"Spectral"

and then

"Spectral_r"

to invert it but there doesn't seem to be a

"Turbo_r"

colormap in Rerun. https://cdn.discordapp.com/attachments/1416441589835763902/1416441590456516709/Screenshot_2025-09-13_at_17.07.40.png?ex=68c6db87&is=68c58a07&hm=064edc1d7d4e781629b2bb916b6a1f0f64b6c895ff2684368310e434b4626247& https://cdn.discordapp.com/attachments/1416441589835763902/1416441590985134091/Screenshot_2025-09-13_at_17.12.53.png?ex=68c6db87&is=68c58a07&hm=841421261ec4b03074883a72084d044f0376960346f8b0508ccf234ef7b4dcac&

Hello, I would like to know if there is a way to create a Graph where I use .with_positions() only on some GraphNodes and the positions of the others are computed using the physic simulation.

Hi Rerun developers! 1. I am Building Rerun-SDK from source using [CMake in Detail article](https://ref.rerun.io/docs/cpp/stable/md__2home_2runner_2work_2rerun_2rerun_2rerun__cpp_2cmake__setup__in__detail.html). I succeeded doing so on Mac and on Ubuntu24.04. My goal is to cross-compile it inside the docker image - it fails on that stage. My steps are as follows (in docker I start from

phase 2

):

sh
#!/bin/zsh

echo ":: phase 1"

mkdir src
cd src

RERUN_URL=https://github.com/rerun-io/rerun/releases/download/0.24.1/rerun_cpp_sdk.zip
RERUN_DIR=rerun_cpp_sdk

wget -q -O ${RERUN_DIR}.zip ${RERUN_URL}
unzip -q -o ${RERUN_DIR}.zip -d .
mv ${RERUN_DIR}/* .
rm -rf ${RERUN_DIR} ${RERUN_DIR}.zip

echo ":: phase 2"

cd ..
mkdir build
cd build

cmake \
    -D CMAKE_INSTALL_PREFIX=/usr/local/ \
    -D CMAKE_BUILD_TYPE=Release \
    -B . \
    -S ../src

cmake --build . --config Release --target rerun_sdk

In docker the main issues are:

sh
fatal error: xsimd/xsimd.hpp: No such file or directory
   25 | #include <xsimd/xsimd.hpp>
      |          ^~~~~~~~~~~~~~~~~
fatal error: mimalloc.h: No such file or directory
   53 | #  include <mimalloc.h>
      |            ^~~~~~~~~~~~

I tried to add paths, but arrow builds using

external project add

and that is why my flags are ignored. 2. I tried to add

arrow-cpp

using

pixi

in Docker, but pixi has only arrow-cpp < 14 from conda-forge repo. I suppose it is incompatible with rerun.

sh
> pixi init
Created /build/pixi.toml
> pixi add arrow-cpp==18.0.0
Error:   x failed to solve requirements of environment 'default' for platform 'linux-64'
  |->   x failed to solve the environment
  |
  `-> Cannot solve the request because of: No candidates were found for arrow-cpp ==18.0.0.

I would greatly appreciate any help/recommendations for that, because I am trying to accomplish this 3 days in a row.

Hi Rerun team, awesome software you guys have built. I am trying to integrate rerun as the visualization and data management layer for an aircraft traffic simulatior I am building with bevy, mostly utilizing its ECS. In the future, I would like to add some egui elements such as buttons and console to send command to the simulation, so researchers don't need to rely on a separate terminal to manage the simulation. I wonder how feasible it is to integrate, for example, a separate panel with egui elements to send commands to the bevy app? Thanks!

Whenever I build my project (simple hello world from https://rerun.io/docs/getting-started/quick-start/cpp ), the arrow_cpp target gets reconfigured, rebuilt and reinstalled even though I never tried to install it (see log below). This takes an unreasonable amount of time for something that is so much part of my dev cycle, it will sadly be a dealbreaker if I cannot find a way around.

>cmake --build build
MSBuild version 17.8.5+b5265ef37 for .NET Framework

  Checking File Globs
  1>Performing update step for 'arrow_cpp'
  -- Already at requested tag: apache-arrow-18.0.0
  No patch step for 'arrow_cpp'
  Performing configure step for 'arrow_cpp'
  -- arrow_cpp configure command succeeded.  See also C:/Users/emichel/SourceCode/SparkleOptimizer3/build/_deps/rerun_sdk-build/arrow/src/arrow_cpp-stamp/a
  rrow_cpp-configure-*.log
  Performing build step for 'arrow_cpp'
  -- arrow_cpp build command succeeded.  See also C:/Users/emichel/SourceCode/SparkleOptimizer3/build/_deps/rerun_sdk-build/arrow/src/arrow_cpp-stamp/arrow
  _cpp-build-*.log
  Performing install step for 'arrow_cpp'
  -- arrow_cpp install command succeeded.  See also C:/Users/emichel/SourceCode/SparkleOptimizer3/build/_deps/rerun_sdk-build/arrow/src/arrow_cpp-stamp/arr
  ow_cpp-install-*.log
  Completed 'arrow_cpp'
  rerun_sdk.vcxproj -> C:\Users\emichel\SourceCode\SparkleOptimizer3\build\_deps\rerun_sdk-build\Debug\rerun_sdk.lib

I'm writing some code that converts data from one format into Rerun types for visualization. I'd like to write unit tests which verify the produced

rerun::Points3D

objects contain identical data to the original objects. I'm finding this challenging, because the

rerun::Points3D

converts the provided

Collection<rerun::components::Position3D>

into an Arrow-encoded array, and never stores an intermediary representation. There don't seem to be accessor functions on the

rerun::Points3D::positions

member, which is a

std::optional<ComponentBatch>

. From what I can see, there isn't a way to convert from the Arrow-encoded array back to the original datatype in the C++ SDK. Is there a way to extract the original data from the Arrow-encoded array?

For some reason in 0.24.1 I am not able to step though the timeline by using the left and right arrow keys. This worked on previous versions like 0.18 and 0.20. I'm not sure exactly on which version arrow keys stopped working to play forwards/backwards. I have tested this behaviour on both a MacBookPro M1 16 inch as well as an Ubuntu 20 machine.