RAW8 Format Handling in SIPL/UDDF on Jetson AGX Thor (L4T R39.2.0)

Hi NVIDIA Team,

We are investigating a RAW8 capture issue using the SIPL/UDDF framework on Jetson AGX Thor running L4T R39.2.0.

Our camera outputs RAW8 data over MIPI CSI-2, with data type 0x2A. However, even though "inputFormat": "raw8" is specified in the SIPL JSON configuration, the resulting capture configuration appears to use PIX_PACK_RAW12 with CSI-2 data type 0x2C and memory format VI_PIXFMT_FORMAT_T_R16.

The same camera works correctly with RAW8 through the V4L2 capture path, where the CSI-2 data type and memory format are configured appropriately.

Could you please clarify whether RAW8 is fully supported in the SIPL/UDDF capture path on this platform? If supported, could you advise how to configure it correctly or whether this behaviour is a known limitation?

Thanks.

hello 230801055,

unfortunately,
RAW8 is not fully integrated into the SIPL/UDDF sensor configuration path although the lower VI capture layer already supports RAW8.

may I double check your sensor types, FYI, JP-7.2 support GMSL and CoE only.

Hi @230801055 and @JerryChang,

Thanks for the clarification. NVIDIA’s response confirms that RAW8 support is incomplete in the SIPL/UDDF sensor-configuration path, despite support in the underlying VI layer.

The R39.2 documentation lists raw8 in the SIPL Query JSON format fields and RAW8 in the GMSL UDDF deserializer configuration. These declarations alone do not establish that RAW8 works throughout the capture path.

To narrow down the reported RAW12 selection, could you share:

  • The sensor model and mode, and whether the connection is direct CSI-2, GMSL, or CoE. For GMSL, please include the serializer/deserializer models and module driver. This would address Jerry’s question about the supported topology.
  • The resolved virtual-channel inputFormat and cfa in the SensorSystemConfig passed to SetPlatformCfg, alongside the relevant JSON entries. The JSON guide also requires ParseDatabase() before ParseJsonFile() on JP 7.2.
  • The exact log or configuration field showing PIX_PACK_RAW12 and 0x2C, identifying whether it describes receiver programming or the incoming stream.

CSI-2 RAW8 uses DT 0x2A, while RAW12 uses 0x2C, as shown in the Linux CSI-2 definitions. The memory surface format VI_PIXFMT_FORMAT_T_R16 alone does not establish the wire data type.

If the resolved configuration is RAW8 but the receiver is programmed for RAW12, that would help localize the discrepancy downstream of JSON resolution. It would still need NVIDIA’s confirmation as the cause of this issue.

@JerryChang , does the acknowledged integration limitation explain this specific RAW12 selection, and is there a supported workaround or planned fix for the poster’s topology on R39.2.0?

Best regards,
Alejandro Godínez
Embedded Software Engineer, RidgeRun
Contact us: support@ridgerun.com
Developers Wiki | Website