Artefact in Libagrus Image

We are seeing some artefacts in the image captured using Libargus.

HW setup:

  1. Xavier NX - Custom Carrier Board
  2. Image sensor->SubLVDS->FPGA->MIPI->NX (Bayer10BG)

SW setup:

  1. Libargus to capture image - YUV420 output. Pipeline created before JP5.1 (using JP5.0.2), hence using YUV420. ISP does demosaicing, CCM, etc…
  2. VIC - Convert YUV420 to RGB
  3. RGB saved as png using OpenCV

Details of the artefact:

  1. We are seeing color patches (8x8 block) in the image when we zoom it and this noise makes our DL model to fail.
  2. We don’t know the cause of the issue.
  3. Also, we notice that the image has some fogging effect. (Have tried defogging in the settings, didn’t help much)

Details of the attachment.

  1. Original Image:
  2. Area of focus.
  3. Artefact image (color patches/Blocks).

Did you confirm the YUV420 output from Argus?

Yes. The output from libargus is YUV420.

Please try adjust the Edge Enhance strength.
You can check by gst-inspect-1.0 nvarguscamerasrc to check ee-mode/ee-strength


Could you use nvargus_nvraw to capture the raw/JPEG to confirm.


@ShaneCCC ,

With V4l2, the captured image is good without any artefacts. I will check edge enhance strength.

hello ts-guna,

had you apply ISP tuning files? may I also know what’s badge property settings in your sensor device tree?

@JerryChang ,

This is our own sensor module and we don’t have ISP tuning file? If this needs ISP tuning, then the ISP in the NX is not all use-able for us as we don’t have access to ISP tuning software. We don’t need true color or any other IQ tuning, we just the need the image without this sort of artefact. Can we do that without the need to go to some NVIDIA ISP-IQ partners?

@ShaneCCC ,

With raw format (direct bayer output) where we apply demosaicing on the application layer, this issue is not seen.

First we need to figure out the root cause to provide possible solution.
Please check the badge info in device tree to avoid load incorrect ISP configure. And check with nvargus_nvraw even you have check raw by captured by v4l2-ctl.

I couldn’t find badge info in the device-tree. With the log from nvargus-deamon, I think it’s not loading any wrong isp file.

Mar 02 13:02:38 ts nvargus-daemon[3073]: LSC: LSC surface is not based on full res!
Mar 02 13:03:06 ts nvargus-daemon[3073]: === nvargus_nvraw[3074]: CameraProvider destroyed (0xffff805c5e10)=== nvargus_nvraw[3074]: Connection closed (FFFF8729E900)=== nva>
Mar 02 13:03:06 ts nvargus-daemon[3073]: NvPclHwGetModuleList: WARNING: Could not map module to ISP config string
Mar 02 13:03:06 ts nvargus-daemon[3073]: NvPclHwGetModuleList: No module data found
Mar 02 13:03:06 ts nvargus-daemon[3073]: OFParserGetVirtualDevice: NVIDIA Camera virtual enumerator not found in proc device-tree
Mar 02 13:03:06 ts nvargus-daemon[3073]: ---- imager: No override file found. ----
Mar 02 13:03:06 ts nvargus-daemon[3073]: LSC: LSC surface is not based on full res!
Mar 02 13:03:06 ts nvargus-daemon[3073]: === nvargus_nvraw[3142]: CameraProvider initialized (0xffff809c3b90)LSC: LSC surface is not based on full res!

The output of

nvargus-nvraw --lps

nvargus_nvraw version 1.14.0
Number of supported sensor entries 1
Entry  Source Mode      Uniquename             Resolution   FR  BitDepth  Mode
Index  Index  Index                                             CSI Dyn   Type
  0      0      0          jakku_front_imx392   1920x1200   22  10  10    Bayer 

Could you provide the nvraw file by nvargus_nvraw.

Image res: 1920x1200 - Bayer BG, 10 bit mode

test.nvraw (4.4 MB)

I would like to check the same scene with the artefact issue, to confirm if nvraw without problem but preview have the artefact.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.