I’m currently testing out a colour IMX296 sensor with Gstreamer and Libargus. When reviewing footage, I noticed that there is oversharpening happening in the frames. Looking through the gstreamer/libargus preferences and settings, I could not find any way of disabling oversharpening. Is there a way to disable or minimize oversharpening with Libargus? For reference, here is the gstreamer pipeline that I used:
please narrow down the issue, you may dump the IMX296 raw images to check capture frames before processed by ISP.
for example,
you may review the sensor pixel formats, $ v4l2-ctl -d /dev/video0 --list-formats-ext
please using the supported resolution and pixel formats to dump the raw files,
i.e. $ v4l2-ctl --set-fmt-video=width=1920,height=1080,pixelformat=RG10 --stream-mmap --stream-count=1 -d /dev/video0 --stream-to=test.raw
there’re some 3rdparty tools to analysis the raw files, such as 7yuv, IrfanView…etc.
thanks
BTW,
are you having other Jetson platforms (such as Jetson-TX2) to reproduce the issue?
had you also contact with your sensor vendor for the tuning supports.
thanks
The sensor pixel format is RG10 for my colour camera and Y10 for my monochrome camera, and when dumping and reviewing RAW videos (I used ffmpeg to encode them) the oversharpening is not present, the image looks much cleaner, more useful for my algorithms and better on the eye. I still believe the issue is due to post-processing done on the Nvidia ISP.
I have a Jetson Nano I could test this on, but not a TX2 or TX2 NX.
thanks for sharing the info, yes, this should related to the post-processing within ISP.
since the software were different for each Jetson series (i.e. Jetson Xavier NX v.s. Jetson Nano),
please also have confirmation on Jetson Nano platform, this may helps to narrow down the issue.
Here is a still image captured with argus_camera (denoising, EE, digital ISP gain disabled). HW setup: Jetson NX devkit, imx477 camera, Edmund Optics 4mm lens.
I don’t see white halo surroundings but lots of noise according to your capture results.
I’ll also arrange resources to check this from our side, thanks
I have tried argus oneshot and obtained the following images (With a Xavier NX and IMX296 sensor). Note how the outdoor image still shows oversharpening (trees and the roof have a white halo).
quick questions,
are you working with sensor vendors to have ISP tuning parameter applied? could you please contact with them to figure out what’s the settings being used.
thanks
I’ve upload a pre-built binary based-on r32.5.1 to disable ISP sharpness function. Topic175200_May27_libnvscf.zip (2.7 MB)
could you please have a try to replace /usr/lib/aarch64-linux-gnu/tegra/libnvscf.so with the attachment.
please also share the results for reference,
thanks
I see, the pre-built binary I’ve just attach is based-on JetPack-4.5.1 / l4t-r32.5.1
here’s another binary based-on JetPack-4.4 / l4t-r32.4.3 for your verification. Topic175200_JP44_libnvscf.zip (2.7 MB)
note,
it just a workaround to disable ISP sharpness function.
please do have verification and share us the results, so we could have investigation this further.
thanks
This time, I was able to capture footage (See attached screenshots). However I do not see any visual changes regarding the oversharpening. In the screenshots the trees and the roof both show white halos.