How do I run a image processing pipeline fast for multiple images processing (concurrently)?

Accelerated Computing Intelligent Video Analytics DeepStream SDK
, , ,
1

Hi,

I’m running a pipeline for processing multiple jpeg images with yolo object detection as below, the pipeline is:

pipeline = “appsrc ! jpegdec ! videoconvert ! nvvideoconvert ! nvv4l2h264enc ! h264parse ! nvv4l2decoder ! nvinfer config-file-path=yolo.txt ! nvvideoconvert ! nvdsosd ! nvvideoconvert ! videoconvert ! jpegenc ! fakesink”

BTW, I’m running this pipeline in python.

The thing is that it works fine sequentially but I had to reset by pipeline.set_state(Gst.State.READY) and pipeline.set_state(Gst.State.PLAYING) once each image processing is done. This takes too much time to handle 60 images in this sequential fashion (each processing takes 0.25 second).

I’ve tried to apply threading.Thread(target=run_pipeline) but it seems like each thread should wait until the previous image processing is done – then Gst.State.READY again…

I could create a function to create pipeline in each thread, but it also takes too much time to create pipeline (0.3 second).

I want to create a pipeline only one time (0.3 second), and
run the pipeline multiple times concurrently (?) or without any delay for processing 60 images in less than 1 second.

Should I set a hardware to do this? or can I set “Gst.State.READY” for multiple images at the same time?

Any solutions / suggestions / comments will be appreciated.
Thanks in advance.
– Sean.

My environments are:

  • Jetson AGX Orin
  • JetPack 5.0.2 (L4T R35.1.0)
  • Deepstream6.1
  • Python 3.8
3

If all the pictures are in the same format, E.G. they are all JPEG images. You may consider to use multifilesrc.

4

Hi,

Actually, I need to give the bytes of image to “appsrc” cause I’m using webservice to receive the image bytes and process this bytes of image in the Deepstream pipeline (as above).

I was looking for “multiappsrc”, similar to “multifilesrc” as you suggested, but “multiappsrc” does not exist.

Is there a way to use “multifilesrc” for the REST protocol to receive the image in bytes?
Or any other suggestions?

Thanks1
-Sean

5

Since you are using appsrc, so you can implement the continuous receiving images with appsrc only.

That means your appsrc implementation has something wrong. Please debugging your appsrc implementation.

6

Got it. Thanks!

7

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.