Segfault running Argus Camera 3.2 with an IMX296 (no IMU), Argus Camera 3.1 works

[rowanborder]

@hemalshahNV @jaiveersinghNV

Similar to #53 and #51, I am also experiencing a segfault when running ISAAC-ROS 3.2 with an IMX296 camera (which does not have an IMU). It appears that ISAAC-ROS 3.2 expects every camera to have an IMU, as the segfault occurs in nvidia::isaac::ArgusCamera::ArgusImpl::fillImuCalibrationData. This is not the case for cameras like the IMX296.

Reverting to ISAAC ROS 3.1 using this fix (https://forums.developer.nvidia.com/t/argus-camera-segfault-on-jetpack-6-2/324192/9) works, but I would like to upgrade to ISAAC ROS 3.2 to use the new hardware timestamping feature.

Thanks in advance for your help resolving this issue!

[sgillen]

Hi @rowanborder, the fillImuCalibrationData shouldn't assume that there is an IMU, but maybe there is a bug on our side.

That function will check a bool in the calibration data from the camera EEPROM to determine if there is an IMU, if it does not, that function should return early without doing anything. Do you happen to know what the EEPROM data looks like from that camera?

[rowanborder]

Hi @sgillen , I've attached the output from v4l2-ctl --all. I tried to find the EEPROM data from the corresponding i2c device in /sys/bus/i2c/devices/9-001a/eeprom, but it doesn't seem to exist. For reference I'm using a ConnectTech Boson Carrier Board with 4x IMX296 cameras (https://connecttech.com/product/boson-for-framos-carrier-board-for-nvidia-jetson-orin-nx/). Thanks!

v4l2-ctl--all_output.txt

[sgillen]

Thank you - this may be the same root cause as #51 - we're working on a fix now.

[Calvin-InDro]

Thank you - this may be the same root cause as #51 - we're working on a fix now.

Following closely, thank you for your efforts @sgillen + team.

[MeticulousFishFrier]

Has there been any updates on this? @sgillen

[sgillen]

Hello, sorry for the delay, was OOO for awhile. Sorry I don't have a fix out yet but we have not forgotten this thread.

[sgillen]

Hello, we just released a fix, it lives in isaac_ros_nitros - (NVIDIA-ISAAC-ROS/isaac_ros_nitros#58). This definitely resolves an issue related to segfaulting and IMUs, please give it a shot and update us.

[josepqp]

hello @sgillen , I have the package installed using apt-get in my jetson, but checking the new release I saw that you just changed the "so" for the new one.

Now the 3.2 version works like the 3.1, no segfault, but I have frame drops like with the 3.1 version, the output looks like:

[component_container-1] 2025-07-11 11:38:56.797 WARN extensions/hawk/components/argus_camera.cpp@778: Failed to get IMU calibration data, skipping fillImuCalibrationData
[component_container-1] 2025-07-11 11:38:56.801 WARN extensions/hawk/components/argus_camera.cpp@778: Failed to get IMU calibration data, skipping fillImuCalibrationData
[component_container-1] 2025-07-11 11:39:00.972 WARN extensions/hawk/components/argus_camera.cpp@1735: Frame drop in module_id 0 camera_id 0
[component_container-1] 2025-07-11 11:39:00.983 WARN extensions/hawk/components/argus_camera.cpp@1735: Frame drop in module_id 1 camera_id 1

The first two messages are right, I don't have IMU and now it's only a warning. But after that I constantly get the other messages of frame drops.

The other thing I notice, I'm getting the images with a hardware trigger (there I have a fps = 1), when I look to the received images I think I'm getting the correct number. Then it looks like this frame drop message is not right.

When I finish the program, on the summary I get things like:

2025-07-11 11:53:19.491 WARN extensions/hawk/components/argus_camera.cpp@1813: Module_id 1, camera_id: 1, frames_dropped 853 drop_percent: 99.882904%

2025-07-11 11:53:26.599 WARN extensions/hawk/components/argus_camera.cpp@1813: Module_id 0, camera_id: 0, frames_dropped 859 drop_percent: 99.883721%

I'm almost sure that I'm not having a 99'9 % of frames dropped. other wise I wouldn't see any update on the images.

[sgillen]

Hi @josepqp - it looks like the segfault is fixed - could you open a new issue to track the faulty warnings?

The problem is argus_camera is not really set up for an external hw trigger, it is assuming your expected framerate is at 30hz, when you are actually hardware triggering at 1hz. That's annoying for you (sorry about that!). One option is to suppress these warnings with GXF_LOG_LEVEL=ERROR