Extracting frame and save to file from nvinfer

Accelerated Computing Intelligent Video Analytics DeepStream SDK
,
1

• Hardware Platform (Jetson / GPU) Javier
• DeepStream Version 6.2

I am trying to save frames from multiple rtsp streams to files. This seems to be similar to the example in deepstream_imagedata-multistream, but I keep getting the following errors:

If I try to attach a probe to the source pad of the source bin like so:

    pad = analytics.get_static_pad("src")
    pad.add_probe(
        Gst.PadProbeType.BUFFER, save_image_probe, 0
    )

Then I am able to read the metadata (using gst_buffer_get_nvds_batch_meta) but I cannot use get_nvds_buf_surface because “Currently we only support RGBA color Format”.

If I try to attach the probe to the tiler, as in the example:

    tiler_sink_pad = tiler.get_static_pad("sink")
    tiler_sink_pad.add_probe(Gst.PadProbeType.BUFFER, save_image_probe, 0)

Then I am not able to read the metadata any more, and get the error: AttributeError: 'NoneType' object has no attribute 'frame_meta_list' - the result of using gst_buffer_get_nvds_batch_meta is None. Could you tell me the correct approach to what I am trying to achieve here? I would like to be able to add a probe which will save the image from each source (not the tiled output) to different files.

Note: the sources/source bin pads are created like this:

Gst.Bin.add(new_bin, uri_decode_bin)
bin_pad = new_bin.add_pad(Gst.GhostPad.new_no_target("src", Gst.PadDirection.SRC))
3

Could you try to add a nvvideoconvert to convert the format to RGBA like the demo deepstream_imagedata-multistream.py.

4

I am already using the nvvideoconvert element. Here is the linked stages of my pipeline:

    streammux.link(streammux_pgie_queue)
    streammux_pgie_queue.link(pgie)
    pgie.link(pgie_tracker_queue)
    pgie_tracker_queue.link(tracker)
    tracker.link(tracker_analytics_queue)
    tracker_analytics_queue.link(analytics)
    analytics.link(analytics_tiler_queue)
    analytics_tiler_queue.link(nvdslogger)
    nvdslogger.link(tiler)
    tiler.link(tiler_conv_queue)
    tiler_conv_queue.link(nvvidconv)
    nvvidconv.link(conv_osd_queue)
    conv_osd_queue.link(nvosd)
    nvosd.link(osd_tee_queue)
    osd_tee_queue.link(output_tee)

As far as I can see the nvvideoconvert is configured the same as in the example. Could it be a problem with the ordering of my pipeline elements? If I add the probe to the tiler (which is after the nvvideoconvert), why am I no longer able to access the frame metadata?

5
  1. You didn’t add the capsfilter in your code:
    caps1 = Gst.Caps.from_string("video/x-raw(memory:NVMM), format=RGBA")
    filter1 = Gst.ElementFactory.make("capsfilter", "filter1")

2.The tiler will destroy the batch.

6

Where should I link the capsfilter? I tried in a few places (after analytics/before tiler/after nvvidconv) but I always get the error TypeError: argument element: Expected Gst.Element, but got gi.repository.Gst.Caps.

Given that I want the probe callback to run for every stream individually, I guess that it needs to go before the tiler (as the probe also needs to go before the tiler)?

7

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

It can resolve the “Currently we only support RGBA color Format” issue you attached.

Yes, you are right.

8

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