Closed Finn2708 closed 2 years ago
Thanks for the report! Our SQA has been able to reproduce the problem and we're qualifying a fix for this. We'll publish it as a hotfix as soon as trials are complete.
Thanks for the quick fix, I will test this on monday!
I can confirm that this fixed my issue. Thanks!
Hello,
I ran into the issue already mentioned here. Full error message is:
I tried on multiple x86_64 machines with different GPUs (RTX2060S, RTX3070) running Ubuntu 20.04 LTS. Docker container is pulled from NVIDIA-ISAAC-ROS/isaac_ros_common (Docker Version 20.10.15).
I started the default pipeline like this:
ros2 run usb_cam usb_cam_node_exe --ros-args --params-file camera_params.yaml
(Webcam, 1280x960 @ 30fps, calibrated)ros2 run isaac_ros_image_proc isaac_ros_image_proc
Before subscribing to the
image_rect
topic, VRAM usage is pretty stable.nvidia-smi
(on the RTX2060S system) returns:Roughly 60 seconds after subscribing to
image_rect
withros2 topic echo /image_rect
, VPI fails to allocate the required memory:On the RTX2060S, the pipeline stops rectifying most frames but publishes the non-rectified frames instead, while on the RTX3070 the pipeline crashes after a few
VPI_ERROR_OUT_OF_MEMORY
messages. I have attached the logs of the RTX2060S setup: isaac_ros_image_proc_95128_1652183865298.log usb_cam_node_exe_95110_1652183858680.logAfter cancelling the pipeline, the memory is freed immediately.
By commenting out the nodes of the provided pipeline one-by-one, I believe the issue stems from the
rectify_node
. I tried pinpointing the issue further, but unfortunately I'm not too familiar with the VPI framework myself.I can provide more info if required, but I'm not sure what would be useful. I also have an old Quadro P2200 available that I haven't tested yet, but would be able to if that were of any help.
Kind regards, Finn