NuRec/UJITSO Geometry Processor not rendering colors in headless streaming mode (Isaac Sim 5.1.0 + 6.0.0-dev2)

[fadisgebrael-prog]

Description

Description

When loading a 3DGRUT-exported USDZ (lightfield format) in headless streaming mode,
the Gaussian Splat renders without colors (white/grey). The flag --/UJITSO/geometry=true
is passed but the UJITSO geometry processor is not loading.

This is a duplicate/related to Discussion #565 but specifically for the runheadless.sh streaming mode.

Steps to reproduce

1. Run Isaac Sim in headless streaming mode:
   LIVESTREAM=1 /isaac-sim/runheadless.sh --/UJITSO/geometry=true -v

2. Import a 3DGRUT lightfield USDZ via File → Import

3. Expected: colored Gaussian Splat rendering

4. Actual: white/grey rendering, no colors

Related

• Discussion NuRec Headless Rendering — UJITSO Geometry Processor Not Loading in Python Mode (Isaac Sim 5.1.0) #565: NuRec Headless Rendering — UJITSO Geometry Processor Not Loading
• Blog post claims it works: https://developer.nvidia.com/blog/reconstruct-a-scene-in-nvidia-isaac-sim-using-only-a-smartphone/

Isaac Sim version

5.1.0 + 6.0.0-dev2

Operating System (OS)

Ubuntu 24.04

GPU Name

Quadro RTX 6000

GPU Driver and CUDA versions

Driver Version: 570.172.08 CUDA Version: 12.8

Logs

No response

Additional information

No response

[michaellin6]

Hi @fadisgebrael-prog,

To narrow this down before looking at the streaming path: the RTX renderer has a silent fallback that exactly matches your symptom (geometry present, colors gone), and a few grep lines on your existing log will tell us whether that's it.

Where the log lives

runheadless.sh writes to ~/.nvidia-omniverse/logs/Kit/Isaac-Sim Full Streaming/<version>/kit_*.log (most recent file per launch). Inside the nvcr.io/nvidia/isaac-sim:5.1.0 container that's /root/.nvidia-omniverse/.... Your -v is enough verbosity.

1. Did --/UJITSO/geometry=true actually take effect?

grep -n "UJITSO/geometry" kit_*.log

If you see skipping initialization as '/UJITSO/geometry'=false, the flag didn't reach the carb store before plugin init and the processor stayed off — that alone explains colorless output. If that line is absent, the processor is up; go to step 2.

2. Which subsystem dropped color?

grep -nE "NuRecFieldAsset::loadAssetData|associated field asset for NuRec Volume|field data for NuRec Volume|NuRecVolume::postSync|ParticleFieldEmissive|GaussianEmissive\.mdl as fallback" kit_*.log

What I most expect to see, given the symptom:

Failed to load ParticleFieldEmissive.mdl, ParticleField Gaussian prims will use GaussianEmissive.mdl as fallback
Failed to finalize ParticleFieldEmissive material, ParticleField Gaussian prims will use GaussianEmissive.mdl as fallback

When that fires, the renderer silently substitutes a single-color emissive MDL — geometry stays, per-splat color is gone. NuRecFieldAsset::loadAssetData hits instead would mean the USDZ field data didn't unpack the way hydra expects.

What to send back

• Output of both greps
• The first ~30 lines of the same log (kit version banner)
• Attaching the full log is best if you can

Also useful: one non-streaming run with a frame writer/capture for comparison — separates "renderer never produced color" from "WebRTC composite dropped it". If your 3DGRUT scene is shareable (or derived from nvidia/PhysicalAI-Autonomous-Vehicles-NuRec), let me know and I'll try to repro locally.

Thanks!

[fadisgebrael-prog]

Thanks for your answer, here's what i found:

The carb.ujitso.processor.geometry.plugin reads /UJITSO/geometry
at t=7.762ms during initialization and always finds it false,
regardless of how the setting is configured.

All failed attempted:

1. --/UJITSO/geometry=true command line flag
2. OMNI_KIT_ARGS_EXTRA=--/UJITSO/geometry=true env var
3. Kit file [settings] section
4. Kit file [settings.persistent] section
5. omni.kit.raycast.query extension.toml changed to true
6. user.config.json with UJITSO geometry=true

Here's the debug log:
kit_debug.log

Thanks a lot in advance