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.

Hey there, I am wondering what's the best way to handle gRPC server restarts on the client side. If I shut down my gRPC server, the client does not throw an exception (Rust) that I could catch [at least nothing that I know of]. It also does not auto-reconnect once the server is up again. It would be great to have a way to figure out that connection got lost and then I can handle it in my code on my own. or even better: auto-reconnect inside rerun SDK, but I take what I can get 🙂

I'm running into an issue with visualizing some rerun results on a newer Windows laptop that I was able to do with my previous Ubuntu Laptop. Here's my setup - I have a workstation and a laptop, with a native viewer open on my laptop; and some code logging to rerun on my workstation using the rerun-sdk. I have ssh-ed into my workstation from my laptop with remoteForwarding enabled on the appropriate port. I previously used to use rr.connect_tcp("0.0.0.0:9876") to visualize things on my local native rerun viewer, but it doesn't seem to be connected now on my windows laptop. (I also tried the newer rr.connect_grpc , but that hasn't worked either). Is someone able to help me figure out why? Some additional things I checked - I made sure widnows is allowing rerun.exe to communicate over the network Im on I've tried out a few different ports / bind addresses for rerun I've tried the same connection on my older Ubuntu laptop (rerun-sdk version 0.20.3 ), and that works; on my windows laptop neither 0.20.3 nor the later 0.24.* work

Hey, I am running a gRPC server using Rust SDK's

serve_grpc_opts

function. I pass

MemoryLimit::parse("2GB")

(also tried with

::from_bytes

and the byte count), but my rerun server process goes wild with memory usage. It basically continues to accumulate memory until the OS kills it or hangs (see screenshot it's already at around 25Gbyte). I run it in a Docker container on Ubuntu, if that matters. The web-viewer recording itself hovers around ~2Gbyte, so that is good. But if the server runs OOM that doesn't help much. Is there something I misunderstand or am missing about the configurable memory limits? On code level, this is all I am doing:

let rec = RecordingStreamBuilder::new("test")
        .recording_id("test-recording")
        .serve_grpc_opts("0.0.0.0", 9876, MemoryLimit::parse("2GB")?)?;

loop {
    std::thread::park();
}

https://cdn.discordapp.com/attachments/1408534249694302229/1408534249858007162/CleanShot_2025-08-22_at_12.25.53.png?ex=68aa173e&is=68a8c5be&hm=4d3f24a0470516d847807d07eab5868c7d21a2a8be971f681b61543a5d73c42b& https://cdn.discordapp.com/attachments/1408534249694302229/1408534250340483142/CleanShot_2025-08-22_at_12.26.022x.png?ex=68aa173e&is=68a8c5be&hm=a266a0240ef24cea32cad8e490a29efd4226c8dd394920136df51301c65c7f83&

I'm trying to create a Transform3D with xyz and rpy (roll, pitch, yaw) data, but I can't find any mention of euler angles in any of the rerun docs. Is there really no easy way to create a Rotation3D object out of euler angles using the rerun API?

Hi, Rerun Team! Firstly, thanks for building awesome software:) In our project, we want to switch from using ROS and rviz to Rerun. Could you please tell if Rerun has an equivalent to ROS topics that can be used as a communication bus between threads?

I know it can be set from the GUI clicking through settings, but it would be nice if we could make this the default for our app so we don't need every user to change this setting on their own.

I'm using v0.24.1 with the web viewer.

LinkAxis.Independent

works well, but

LinkAxis.LinkGlobal

doesn't log any data. Attached a screenshot for Independent vs Global link. Is there anything wrong with the way I am logging the data? Here's how I have the data logged: For each entity: - Fetch timestamps and values using fetch_data(entity_path, start_time, end_time) - Send data to Rerun:

rr.send_columns(
            entity_path,
            indexes=[rr.TimeColumn("time", timestamp=timestamps)],
            columns=rr.Scalars.columns(scalars=values),
        )

Build a list of views: - For each panel:

rrb.TimeSeriesView(
            origin="/",
            name=panel["name"],
            contents=entity_paths,
            axis_x=rrb.archetypes.TimeAxis(
                link=rrb.components.LinkAxis.LinkGlobal
            ),
        )

Arrange views into a grid layout:

grid = rrb.Grid(*views, grid_columns=1)

Send the blueprint to Rerun:

rr.send_blueprint(
        rrb.Blueprint(
            grid,
            collapse_panels=True,
        ),
        make_active=True,
    )

https://cdn.discordapp.com/attachments/1405995563413143694/1405995564037968035/Screenshot_2025-08-15_at_3.22.02_PM.png?ex=68a0dae8&is=689f8968&hm=164bd9c87ac6d25e2db3d7de055eab4b591ce43b43f1429e9c3f6b466ba0ff8f& https://cdn.discordapp.com/attachments/1405995563413143694/1405995564444946473/Screenshot_2025-08-15_at_3.24.03_PM.png?ex=68a0dae8&is=689f8968&hm=89724618fee1ed6db1659b575c4df51270ba77b34865be98b282ae792325e7ae&

Is there a tutorial for logging with rerun on a remote machine, which I am SSHd to with port forwarding, and viewing the data on my local machine? There are lots of random questions about it but I still can't make it work. Would appreciate a clean working example hosted on the rerun docs! Right now I reverted to just trying to get the web viewer to work locally (no remote) with the code provided on the website

"""Demonstrates how to log data to a gRPC server and connect the web viewer to it."""

import time

import rerun as rr

rr.init("rerun_example_serve_web_viewer")
# Start a gRPC server and use it as log sink.
server_uri = rr.serve_grpc()

# Connect the web viewer to the gRPC server and open it in the browser
rr.serve_web_viewer(connect_to=server_uri)

# Log some data to the gRPC server.
rr.log("data", rr.Boxes3D(half_sizes=[2.0, 2.0, 1.0]))

# Keep server running. If we cancel it too early, data may never arrive in the browser.
try:
    while True:
        time.sleep(1)
except KeyboardInterrupt:
    print("\nShutting down server...")

However no data shows up in the web viewer and it gives me the error "Data source rerun+http://127.0.0.1:9876/proxy has left unexpectedly: gRPC error, message: "js api error: TypeError: NetworkError when attempting to fetch resource.""

Hi Rerun team Firstly, thanks for building awesome software! I'm logging three camera feeds (640x480, RGB+depth) and 13 piper robot proprioception fields per step to rerun. I'm running at 20Hz, so I have a budget of 50ms for each step. The below logging code in rerun takes around 30ms, sometimes jumping above 50ms, and is reducing my FPS. Am I doing anything naive here, and is there a way to speed up the

rr.log

calls below? Thanks for your help! Tim

python

def _log_to_rerun(payload: tuple) -> None:

  rec, seq, rgb_items, depth_items, proprio_items = payload

  rr.set_time(
      timeline="episode_step",
      sequence=seq,
      recording=rec,
  )

  for path, frame in rgb_items:
    rr.log(
        path,
        rr.Image(frame),
        recording=rec,
    )

  for path, frame in depth_items:
    rr.log(
        path,
        rr.DepthImage(frame, meter=1000.0, depth_range=(0.0, 1500.0)),
        recording=rec,
    )

  for prop_key, proprio_frame in proprio_items:
    for dim in range(proprio_frame.shape[-1]):
      rr.log(
          f"{prop_key}/d{dim}",
          rr.Scalars(proprio_frame[..., dim]),
          recording=rec,
      )

Using a windows machine to remote connect to a jetson board, worked normal when openning the web-viewer page on browser once forwarded the port 9090, but could not get any response but the welcome page. Codes below:

import rerun as rr

rr.init("test_app", spawn=False)

# 1. Start the gRPC and web servers
grpc_uri = rr.serve_grpc()
rr.serve_web_viewer(connect_to=grpc_uri)

# 2. Log the data (this adds it to a buffer)
rr.log("my_points", rr.Points2D([[0, 0], [1, 1], [2, 2]]))

What did I miss here?

Hi all, I'm trying to stream from my Pi 5 to a web viewer that I can view on another machine. Things work okay, except the only way I can connect to the viewer and see data from my remote machine is if I use the following URL exactly:

http://<SERVER_IP>:9090/?url=rerun%2Bhttp%3A%2F%2F<SERVER_IP>%3A9876%2Fproxy

. The only way I found this URL is from the CLI when I used

--serve-web

. Using

http://<SERVER_IP>:9090/?url=rerun+http://<SERVER_IP>:9876/proxy

works in a browser on the pi using

localhost

but I can't seem to get it to work in a browser on my remote machine without using the URL above. Is there a more generic URL I can use? I've been looking through the docs with no success. Thanks in advance!

Hi All, i am using the visualiser with both Python and C++ SDK, using pixi. => rerun-sdk = ">=0.23.4,<0.24" The python logger works just fine to the visualisation, but the C++ code gives the following error in the visualiser: "dropping LogMsg due to failed decode: Failed to read chunk: Bad chunk schema: Missing row_id column". Maybe we are doing something wrong, the 0.24 release is not yet released for pixi so i can't just upgrade. C++ snipet:

// Create the RecordingStream for rerun.
recording_stream_ = std::make_shared<rerun::RecordingStream>("DataProcessor");

// Spawn a viewer.
recording_stream_->spawn().exit_on_failure();
recording_stream_->log("Input Image",

rerun::Image(rerun::borrow(cv_input_rgb.data, cv_input_rgb.total() * cv_input_rgb.elemSize()),
  rerun::WidthHeight(static_cast<uint32_t>(cv_input_rgb.cols),
  static_cast<uint32_t>(cv_input_rgb.rows)),
  rerun::ColorModel::RGB));

.

I'm running hard up against the gRPC broadcast receiver lagging behind until it errors out. Wondering if I'm holding it wrong or something of that nature

2025-07-31T08:16:52.893796772Z ERROR re_grpc_server: Error reading message from broadcast channel: channel lagged by 281974

Minimal repro: https://gist.github.com/hansihe/522103ad9c5d4e55e9bcfd99213f5b80 I can push a full crate to github too if that makes things easier to dig into, just let me know. I am attempting to use

re_renderer

directly within a custom

egui

application (using

wgpu

). With nothing queued in

ViewBuilder

I get nothing and see the background as expected. When I

queue_draw

a

GenericSkyboxDrawData

, I get a fully black viewport. Behavior of other renderers varies, but I never get any meaningful output. I captured a frame in RenderDoc to try to get more insight into what might be happening. The black screen seems to stem from the color value in the shader `nan`ing. Tracing it back in the debugger it seems to originate from the

re_renderer

frame uniform buffer

which is bound to descriptor set 0 being filled with nulls only. According to RenderDoc the buffer is only `clear`ed and barriered before it is used in the shaders for the

re_renderer

renderers. I imagine I am probably missing some important step in my code, does someone know what I might be doing wrong?

I am receiving a h264 video stream from a UDP socket and I would like to log it with the new VideoStream in C++. More specifically, I am receiving UDP packets (of length 1316 or a smaller multiple of 188) and send them to rerun with:

rec.log("video_stream",rerun::archetypes::VideoStream{}.with_sample(rerun::components::VideoSample{data}));

I had a first error similar to the one mentioned here: https://discord.com/channels/1062300748202921994/1397335975364919392 I applied the

uint32_t

fix and the next error was the ffmpeg version. Got the recommended version and the video stream doesn't get displayed:

Failed to decode video: FFmpeg error: [h264] [error] error while decoding MB 2 64, bytestream 16292

It gets the Dimensions of the video stream right (1920x1080) and the sent data chunks appear in the Video samples table but no video gets displayed. The video stream is created from that command:

ffmpeg -f rawvideo -pixel_format rgba -video_size {width}x{height} -framerate {frameRate} -i - -vf vflip -c:v libx264 -preset veryfast -tune zerolatency -f mpegts udp://localhost:1234?pkt_size=1316

and it is displayed correctly with

ffplay

.

Hello, I want to log data from two different sources (a python script and a c++ executable) in the same rerun recording. I would like the data to be saved to file as I log it with rerun (i.e. not all al the end) and also to be able to visualize it live in a single viewer. Is it possible with rerun? How would I implement this?

Hello, I have a list of joints from a robot encoder and I want to log them with rerun. Is it possible to log all of them in one call without looping on the list and call rr.log for each value?

I'm trying to render a surface patch generated from HiRISE DTM and orthoimages (e.g. https://www.uahirise.org/dtm/dtm.php?ID=PSP_002387_1985). I used the HiRISE DTM Importer blender plugin (https://hemelmechanica.nl/hirise-docs/quickstart.html) to generate the mesh from the DTM image and texture it use a matching grayscale orthoimage. Everything looks correct in Blender and also in https://gltf.report/ and https://gltf-viewer.donmccurdy.com/. However, in rerun, the model looks bright red as if its texture has the greyscale image in the R channel but G and B are just set to 0. Changing AlbedoFactor to rgb(255,0,0) has no effect on how the surface patch looks. But changing it to rgb(0,255,0) or rgb(0,0,255) or rgb(0,255,255) just makes the entire surface patch pitch black. I'm not sure what I'm doing wrong since when I import other GLB files in rerun (e.g. a glider model I found and a free rubber ducky model I found to use as a test), they look normal. https://cdn.discordapp.com/attachments/1398520508022132866/1398520508907126846/Screenshot_2025-07-25_at_9.01.37_PM.png?ex=6885a938&is=688457b8&hm=2de6e6ff4fe3488eae2d1a41a87b80ce707f6683f76b8bb3dcbe70be27ebf2df& https://cdn.discordapp.com/attachments/1398520508022132866/1398520509552918639/Screenshot_2025-07-25_at_9.02.07_PM.png?ex=6885a938&is=688457b8&hm=ce2b4e462c1542db2b5c763923f96d77cffd7df42356d0a8e433063d318009b0& https://cdn.discordapp.com/attachments/1398520508022132866/1398520509964095608/Screenshot_2025-07-25_at_9.03.30_PM.png?ex=6885a938&is=688457b8&hm=9c3609eac08dcf3c2f3a5568d8f32793ff84ccc0ec46fde8cbf400ef2b67d991& https://cdn.discordapp.com/attachments/1398520508022132866/1398520510509350963/Screenshot_2025-07-25_at_9.15.54_PM.png?ex=6885a938&is=688457b8&hm=7c68c1b8c144dbef323ffce38a4a1c8a000a248cdeb7f24035c439baaea3187a&

Is there a way I can programmatically set the viewpoint for a Spatial3DView using the PythonAPI so that I don't have to manually pan/zoom/rotate a pointcloud with the mouse everytime to see the view I need?

Hi, I'm trying to get the VideoStream support working on C++. The example on python works correctly, but when trying to port this over to C++ I keep getting:

re_viewer_context::component_ui_registry] Failed to deserialize component of type rerun.components.VideoCodec: Failed to deserialize "rerun.components.VideoCodec"

I've put the RUST_LOG up to debug on both the sender and the receiver but it gives me no other information other than the deserialization error. Any idea what I might be doing wrong? I've attached the code I'm working with. This also dumps the h264 data to a file which does reproduce correctly when using ffplay so I don't think this is an encoding error https://cdn.discordapp.com/attachments/1397335975364919392/1397335975645941882/on-the-fly.cpp?ex=688b3d49&is=6889ebc9&hm=f06d6a86384a371ebdcb0a5c7b02a7556716b8bace7b699e20bee65ea0ef7161&

I tested this by creating an application in C++ using the API: - Reads a blueprint (.rbl file) - Creates a recording stream - Attempts to spawn the viewer - Logs some data. When I compile and run this application, it normally spawns the viewer and functions as expected. But when I run this same application as a root user (using sudo), it gets stuck at reading the blueprint stage itself, and cannot proceed forward, unable to spawn the Viewer. If in my application, a root access is required to be used for other purposes, are there any environment variables I need to pass so that the GUI (Rerun viewer) is able to spawn correctly? Supporting evidence is attached. https://cdn.discordapp.com/attachments/1397186188128817222/1397188114430038036/image.png?ex=6880d054&is=687f7ed4&hm=504cba18bd88a0bb8610bcda35b82d9d1ab99b0acc0ef1e55a0d6ef779b8ee93& https://cdn.discordapp.com/attachments/1397186188128817222/1397188114937544834/image.png?ex=6880d055&is=687f7ed5&hm=9142e882a49cf4ddbe87f89c795155c5dad7fb6053aad5e9ba9c0af738984b6b&

When we make the pointcloud panel vertically small (wide and short panel) the points at either end of the viewer become abnormally big. This goes away when the viewer's aspect ratio is more 1:1 (square). Please let me know if this is a bug or configuration issue? I'm using python rerun-sdk 0.24.0 You can reproduce this even on the rerun nuscenes example: https://rerun.io/viewer?url=https://app.rerun.io/version/0.24.0/examples/nuscenes_dataset.rrd All you'll have to do is vertically squish the lidar panel and zoom in so that there are points in the left and right edges of the lidar viewer. https://cdn.discordapp.com/attachments/1397008726090580112/1397008727021850736/image.png?ex=68802943&is=687ed7c3&hm=15c74c5fb58d00906ea87ae9690e4803b87552e1e5195678cdc72bd501ba5266& https://cdn.discordapp.com/attachments/1397008726090580112/1397008728321949756/image.png?ex=68802943&is=687ed7c3&hm=ab2e9ce469b04d766f5e39418c86389a13667c7cc527f5bfa747cf56a3302fd5&