VideoCapture.read() hangs, nvargus error

Autonomous Machines Jetson & Embedded Systems Jetson Nano
#1

Hi,

On my Jetson Nano I’m trying to do a simple face detection by using a loop in which I read a frame (cap.read(), with cap being an instance of VideoCapture using the gstreamer pipeline), then pass it through a network, display the result and continue looping. After a while (sometimes 15 minutes, sometimes 5 seconds), the script just hangs at cap.read() (actually during the .grab() call).
In the syslog, the error is:

nvargus-daemon[24528]: (Argus) Error InvalidState: (propagating from src/api/ScfCaptureThread.cpp, function run(), line 109)
nvargus-daemon[24528]: SCF: Error Invalidstate: Session has suffered from a critical failure (in src/api/Session.cpp, function capture(), line 667)

I did try a stress test, can’t reproduce this. I do have the last version of Jetpack.
The camera is the Raspberry Pi’s one.

Thanks in advance!

Baptiste

#2

What’s the pipeline?
Any error kernel message?
Have you try any others sensor mode?

#3

Hi,

Here’s the pipeline:

def gstreamer_pipeline(capture_width=1260, capture_height=640, framerate=120, flip_method=0):
    return f'nvarguscamerasrc ! video/x-raw(memory:NVMM), width=(int){capture_width}, ' \
        f'height=(int){capture_height}, format=(string)NV12, framerate=(fraction){framerate}/1 ! nvvidconv ' \
        f'flip-method={flip_method} ! video/x-raw, width=(int)840, height=(int)420, ' \
        f'format=(string)BGRx ! videoconvert ! video/x-raw, format=(string)BGR ! appsink max-buffers=1 drop=true'

I don’t have any kernel message…

What do you mean by other sensor modes?

#4

Check for the sensor mode cap by v4l2-ctl --list-formats-ext and replace the capture_width, capture_height, framerate to any others to try.

#5

Oh ok, yes I tried with different modes, doesn’t change anything…

#6

OK, then did you try the gst-launch-1.0 pipeline like below.

gst-launch-1.0 nvarguscamerasrc ! "video/x-raw(memory:NVMM), width=(int)1280, height=(int)720, framerate=(fraction)120/1" ! nvvidconv flip-method=0 ! "video/x-raw, width=(int)840, height=(int)420, format=(string)BGRx" ! videoconvert ! "video/x-raw, format=(string)BGR" ! fakesink
#7

It doesn’t show the camera.
The last line in the terminal is:

CONSUMER: Producer has connected; continuing.

But no camera showing.

#8

Yes, because it’s fakesink. Have below sink to have the preview.

gst-launch-1.0 nvarguscamerasrc ! "video/x-raw(memory:NVMM), width=(int)1280, height=(int)720, framerate=(fraction)120/1" ! nvvidconv flip-method=0 ! "video/x-raw, width=(int)840, height=(int)420" ! videoconvert ! xvimagesink
#9

Alright, this runs. But it does not seem to crash for now.

#10

Maybe the problem comes from the fact that we ask too much to the cpu/gpu at every iteration of the loop?

#11

You have to narrow down which part cause the problem from your source.

#12

I already did, it’s the

cap.read()

that hangs.

#13

Have a check with below link on your device.

https://devtalk.nvidia.com/default/topic/1025356/jetson-tx2/how-to-capture-and-display-camera-video-with-python-on-jetson-tx2/post/5215101/#5215101

#14

I’m using a Jetson Nano by the way.
I’ll try, but it also uses the cap.read() so the error should be the same.

#15

Same problem here when we use our framerate of 120. However, it seems not to hang with a lower framerate.

#16

Alright, so we tested different configurations but in the end it still crashes after some time… Any other idea?

#17

What kind of the configure will crashes?
Did you have a try boost the system by below commnad

sudo su
nvpmodel -m 0
jetson_clocks

#18

I mean we tried different gstreamer pipelines, but it always crashes…

Yeah sure, we tried this mode, still the same problem.

#19

Hi, any news of the matter…?

#20

I have run the tegra_camera.py on my nano with imx219 PI v2 without any problem.