Image pipeline does not reach paused state

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

• Hardware Platform: GPU
• DeepStream Version: 5.0.0
• TensorRT Version: 7.0.0.11
• NVIDIA GPU Driver Version (valid for GPU only): 460.32.03

Following the deepstream_image_decode_app I’m trying to run a pipeline which takes just an image and eventually should do inferences. However for some reason my pipeline is not even reaching the paused state. I deconstructed the pipeline to give a minimal not working example. Please check my pipeline:

gst-launch-1.0 filesrc location=in.jpg ! jpegparse ! nvv4l2decoder ! m.sink_0 nvstreammux name=m width=1920 height=1080 batch-size=1 ! nvvideoconvert ! jpegenc ! filesink location=out.jpg

When watching the graph for the state change from null to ready I observe that nvvideconvert tries to feed video/x-raw(memory:NVMM) into jpegenc which it’s not capable of. I was wondering if the respective pads shouldn’t negotiate to simple video/x-raw caps? However, even when using a capsfilter in between those elements and doesn’t help it.

Any idea what might be the issue?

3

The command line can run successfully on my T4 device.

out
out2257×614 46.7 KB

4

For me it’s like that. I hit ctrl+c after it get stuck on Pipeline is PREROLLING after a while. Can I supply you with some more debug information?

image
image1510×648 34.4 KB

5

Can you upload the jpg file? Seems the file cause the problem.

6

You may try software decoder instead.
gst-launch-1.0 filesrc location=in.jpg ! jpegparse ! jpegdec ! nvvideoconvert ! ‘video/x-raw(memory:NVMM),format=NV12’ ! m.sink_0 nvstreammux name=m width=1920 height=1080 batch-size=1 ! nvvideoconvert ! jpegenc ! filesink location=out.jpg

7

needed to add nvbuf-memory-type=0 to nvstreammux to make it work