Deepstream+PyQt5+OpenCV - RTSP stream delay

Setup:
• Jetson AGX Xavier
• DeepStream 5.0
• JetPack Version 4.4

Hi,
I’ve been trying to combine the DS python multistream example together with open CV and PyQt5. I’ve also added the tracker from a different example together with its config file into the pipeline.
So without the part where I save the image and send it to the appropriate QLabel in my UI, everything runs smoothly. But as soon as I add in a cv.resize and set the label as an image, a delay is present, which gradually gets worse and worse. My goal is to send frames from 3 different streams into 3 different QLabels but the resulting delay is much worse so I need to at least have 1 stream working properly.
This is the part where I grab the image and send it to the UI inside the tiler_sink_pad function:

def display_frame(self, flag, buff, b_id, obj_meta):
rgbImage=pyds.get_nvds_buf_surface(hash(buff),b_id)
#convert python array into numy array format.
frame_image=np.array(rgbImage,copy=True,order=‘C’)
if flag:
frame_image=self.draw_bounding_boxes(frame_image,obj_meta,obj_meta.confidence)
frame_image1 = cv2.resize(frame_image, (self.dim1), interpolation = cv2.INTER_AREA)
frame_image2 = cv2.resize(frame_image, (self.dim2), interpolation = cv2.INTER_AREA)
frame_image3 = cv2.resize(frame_image, (self.dim3), interpolation = cv2.INTER_AREA)

    self.p1=qimage2ndarray.array2qimage(frame_image1)
    self.p2=qimage2ndarray.array2qimage(frame_image2)
    self.p3=qimage2ndarray.array2qimage(frame_image3)
    
    self.label.setPixmap(QPixmap.fromImage(self.p1))
    self.labels1.setPixmap(QPixmap.fromImage(self.p2))
    self.labels2.setPixmap(QPixmap.fromImage(self.p3))

I’ve tried changing the interval in the config file to 30, skipping frames, setting power usage to max + jetson clocks, but to no avail.

Any information as to why this is happening and possible solutions are welcome. I thought of resetting the pipeline every 200 frames but setting the gst.state to “Null” or “Pause” and then “Playing” again just crashes the whole app.

The entire app(without our proprietary code) :

fps_streams={}
frame_count={}
saved_count={}
global PGIE_CLASS_ID_VEHICLE
PGIE_CLASS_ID_VEHICLE=0
global PGIE_CLASS_ID_PERSON
PGIE_CLASS_ID_PERSON=2
MAX_DISPLAY_LEN=64
PGIE_CLASS_ID_VEHICLE = 0
PGIE_CLASS_ID_BICYCLE = 1
PGIE_CLASS_ID_PERSON = 2
PGIE_CLASS_ID_ROADSIGN = 3
MUXER_OUTPUT_WIDTH=1920
MUXER_OUTPUT_HEIGHT=1080
MUXER_BATCH_TIMEOUT_USEC=4000000
TILED_OUTPUT_WIDTH=1920
TILED_OUTPUT_HEIGHT=1080
GST_CAPS_FEATURES_NVMM=“memory:NVMM”
pgie_classes_str= [“Vehicle”, “TwoWheeler”, “Person”,“RoadSign”]

class Deep(QMainWindow):

def __init__(self):
    self.main()
    
    
# tiler_sink_pad_buffer_probe  will extract metadata received on tiler src pad
# and update params for drawing rectangle, object information etc.
def tiler_sink_pad_buffer_probe(self,pad,info,u_data):
    
    # Get dashboard weather data
    start = timer()
    current = timer()
    real = current - self.timed
    self.set_dash(real)
    '''
    # Advance progress bar
  
    '''
    j=1
    frame_number=0
    num_rects=0
    gst_buffer = info.get_buffer()
    if not gst_buffer:
        print("Unable to get GstBuffer ")
        return

    # Retrieve batch metadata from the gst_buffer
    # Note that pyds.gst_buffer_get_nvds_batch_meta() expects the
    # C address of gst_buffer as input, which is obtained with hash(gst_buffer)
    batch_meta = pyds.gst_buffer_get_nvds_batch_meta(hash(gst_buffer))

    l_frame = batch_meta.frame_meta_list
    while l_frame is not None:
        try:
            # Note that l_frame.data needs a cast to pyds.NvDsFrameMeta
            # The casting is done by pyds.NvDsFrameMeta.cast()
            # The casting also keeps ownership of the underlying memory
            # in the C code, so the Python garbage collector will leave
            # it alone.
            frame_meta = pyds.NvDsFrameMeta.cast(l_frame.data)
        except StopIteration:
            break
            print("stopped")

        frame_number=frame_meta.frame_num
        l_obj=frame_meta.obj_meta_list
        num_rects = frame_meta.num_obj_meta
        is_first_obj = True
        save_image = False
        obj_counter = {
        PGIE_CLASS_ID_VEHICLE:0,
        PGIE_CLASS_ID_PERSON:0,
        PGIE_CLASS_ID_BICYCLE:0,
        PGIE_CLASS_ID_ROADSIGN:0
        }
        
        while l_obj is not None:
            
            try: 
                # Casting l_obj.data to pyds.NvDsObjectMeta
                obj_meta=pyds.NvDsObjectMeta.cast(l_obj.data)
            except StopIteration:
                print("stopped")
                #break
            obj_counter[obj_meta.class_id] += 1
            # Periodically check for objects with borderline confidence value that may be false positive detections.
            # If such detections are found, annoate the frame with bboxes and confidence value.
            # Save the annotated frame to file.
            if((saved_count["stream_"+str(frame_meta.pad_index)]%30==0) and (obj_meta.confidence>=0.3 and obj_meta.confidence<0.31)):
                if is_first_obj:
                    is_first_obj = False
                    # Getting Image data using nvbufsurface
                    # the input should be address of buffer and batch_id
                    
                    #convert python array into numy array format.
                    frame_image=np.array(n_frame,copy=True,order='C')
                    #covert the array into cv2 default color format
                    frame_image=cv2.cvtColor(frame_image,cv2.COLOR_RGBA2BGRA)
                    
                
                save_image = True
                j=0
            self.display_frame(1, gst_buffer, frame_meta.batch_id,obj_meta)
            try: 
                l_obj=l_obj.next
            except StopIteration:
                break
        else:
            self.display_frame(0, gst_buffer, frame_meta.batch_id,0)
        # Get frame rate through this probe
        fps_streams["stream{0}".format(frame_meta.pad_index)].get_fps()
        saved_count["stream_"+str(frame_meta.pad_index)]+=1        
        try:
            l_frame=l_frame.next
        except StopIteration:
            break

            
    self.frame_skip+=1
       
    return Gst.PadProbeReturn.OK

def draw_bounding_boxes(self,image,obj_meta,confidence):
    confidence='{0:.2f}'.format(confidence)
    rect_params=obj_meta.rect_params
    top=int(rect_params.top)
    left=int(rect_params.left)
    width=int(rect_params.width)
    height=int(rect_params.height)
    obj_name=pgie_classes_str[obj_meta.class_id]
    image=cv2.rectangle(image,(left,top),(left+width,top+height),(0,0,255,0),2)
    # Note that on some systems cv2.putText erroneously draws horizontal lines across the image
    image=cv2.putText(image,',ID = '+str(obj_meta.object_id),(left-10,top-10),cv2.FONT_HERSHEY_SIMPLEX,0.5,(0,0,255,0),2)
    
    return image
def display_frame(self, flag, buff, b_id, obj_meta):
    rgbImage=pyds.get_nvds_buf_surface(hash(buff),b_id)
    #convert python array into numy array format.
    frame_image=np.array(rgbImage,copy=True,order='C')
    if flag:
        frame_image=self.draw_bounding_boxes(frame_image,obj_meta,obj_meta.confidence)         
    frame_image1 = cv2.resize(frame_image, (self.dim1), interpolation = cv2.INTER_AREA)
    frame_image2 = cv2.resize(frame_image, (self.dim2), interpolation = cv2.INTER_AREA)
    frame_image3 = cv2.resize(frame_image, (self.dim3), interpolation = cv2.INTER_AREA)
    
    self.p1=qimage2ndarray.array2qimage(frame_image1)
    self.p2=qimage2ndarray.array2qimage(frame_image2)
    self.p3=qimage2ndarray.array2qimage(frame_image3)
    
    self.label.setPixmap(QPixmap.fromImage(self.p1))
    self.labels1.setPixmap(QPixmap.fromImage(self.p2))
    self.labels2.setPixmap(QPixmap.fromImage(self.p3))
    
def cb_newpad(self,decodebin, decoder_src_pad,data):
    print("In cb_newpad\n")
    caps=decoder_src_pad.get_current_caps()
    gststruct=caps.get_structure(0)
    gstname=gststruct.get_name()
    source_bin=data
    features=caps.get_features(0)

    # Need to check if the pad created by the decodebin is for video and not
    # audio.
    if(gstname.find("video")!=-1):
        # Link the decodebin pad only if decodebin has picked nvidia
        # decoder plugin nvdec_*. We do this by checking if the pad caps contain
        # NVMM memory features.
        if features.contains("memory:NVMM"):
            # Get the source bin ghost pad
            bin_ghost_pad=source_bin.get_static_pad("src")
            if not bin_ghost_pad.set_target(decoder_src_pad):
                sys.stderr.write("Failed to link decoder src pad to source bin ghost pad\n")
        else:
            sys.stderr.write(" Error: Decodebin did not pick nvidia decoder plugin.\n")

def decodebin_child_added(self,child_proxy,Object,name,user_data):
    print("Decodebin child added:", name, "\n")
    if(name.find("decodebin") != -1):
        Object.connect("child-added",self.decodebin_child_added,user_data)   
    if(is_aarch64() and name.find("nvv4l2decoder") != -1):
        print("Seting bufapi_version\n")
        Object.set_property("bufapi-version",True)

def create_source_bin(self,index,uri):
    print("Creating source bin")

    # Create a source GstBin to abstract this bin's content from the rest of the
    # pipeline
    bin_name="source-bin-%02d" %index
    print(bin_name)
    nbin=Gst.Bin.new(bin_name)
    if not nbin:
        sys.stderr.write(" Unable to create source bin \n")

    # Source element for reading from the uri.
    # We will use decodebin and let it figure out the container format of the
    # stream and the codec and plug the appropriate demux and decode plugins.
    uri_decode_bin=Gst.ElementFactory.make("uridecodebin", "uri-decode-bin")
    if not uri_decode_bin:
        sys.stderr.write(" Unable to create uri decode bin \n")
    # We set the input uri to the source element
    uri_decode_bin.set_property("uri",uri)
    # Connect to the "pad-added" signal of the decodebin which generates a
    # callback once a new pad for raw data has beed created by the decodebin
    uri_decode_bin.connect("pad-added",self.cb_newpad,nbin)
    uri_decode_bin.connect("child-added",self.decodebin_child_added,nbin)

    # We need to create a ghost pad for the source bin which will act as a proxy
    # for the video decoder src pad. The ghost pad will not have a target right
    # now. Once the decode bin creates the video decoder and generates the
    # cb_newpad callback, we will set the ghost pad target to the video decoder
    # src pad.
    Gst.Bin.add(nbin,uri_decode_bin)
    bin_pad=nbin.add_pad(Gst.GhostPad.new_no_target("src",Gst.PadDirection.SRC))
    if not bin_pad:
        sys.stderr.write(" Failed to add ghost pad in source bin \n")
        return None
    return nbin

def main(self):
    # Check input arguments

    args = ['deepstream_imagedata-multistream.py',"rtsp://gadiadmin:gadiadmin@10.0.0.7:554/stream1", 'frames22']

    #if len(args) < 2:
    #    sys.stderr.write("usage: %s <uri1> [uri2] ... [uriN] <folder to save frames>\n" % args[0])
    #    sys.exit(1)

    for i in range(0,len(args)-2):
        fps_streams["stream{0}".format(i)]=GETFPS(i)
    number_sources=len(args)-2

    #global folder_name
    #folder_name=args[-1]
    #if path.exists(folder_name):
    #    sys.stderr.write("The output folder %s already exists. Please remove it first.\n" % folder_name)
    #    sys.exit(1)

    #os.mkdir(folder_name)
    #print("Frames will be saved in ",folder_name)
    # Standard GStreamer initialization
    GObject.threads_init()
    Gst.init(None)

    # Create gstreamer elements */
    # Create Pipeline element that will form a connection of other elements
    print("Creating Pipeline \n ")
    self.pipeline = Gst.Pipeline()
    is_live = False

    if not self.pipeline:
        sys.stderr.write(" Unable to create Pipeline \n")
    print("Creating streamux \n ")

    # Create nvstreammux instance to form batches from one or more sources.
    streammux = Gst.ElementFactory.make("nvstreammux", "Stream-muxer")
    if not streammux:
        sys.stderr.write(" Unable to create NvStreamMux \n")

    self.pipeline.add(streammux)
    for i in range(number_sources):
    #    os.mkdir(folder_name+"/stream_"+str(i))
        frame_count["stream_"+str(i)]=0
        saved_count["stream_"+str(i)]=0
        print("Creating source_bin ",i," \n ")
        uri_name=args[i+1]
        if uri_name.find("rtsp://") == 0 :
            is_live = True
        source_bin=self.create_source_bin(i, uri_name)
        if not source_bin:
            sys.stderr.write("Unable to create source bin \n")
        self.pipeline.add(source_bin)
        padname="sink_%u" %i
        sinkpad= streammux.get_request_pad(padname) 
        if not sinkpad:
            sys.stderr.write("Unable to create sink pad bin \n")
        srcpad=source_bin.get_static_pad("src")
        if not srcpad:
            sys.stderr.write("Unable to create src pad bin \n")
        srcpad.link(sinkpad)
    print("Creating Pgie \n ")
    pgie = Gst.ElementFactory.make("nvinfer", "primary-inference")
    if not pgie:
        sys.stderr.write(" Unable to create pgie \n")
    # Tracker    
    tracker = Gst.ElementFactory.make("nvtracker", "tracker")
    if not tracker:
        sys.stderr.write(" Unable to create tracker \n")
    # Add nvvidconv1 and filter1 to convert the frames to RGBA
    # which is easier to work with in Python.
    sgie1 = Gst.ElementFactory.make("nvinfer", "secondary1-nvinference-engine")
    if not sgie1:
        sys.stderr.write(" Unable to make sgie1 \n")
    print("Creating nvvidconv1 \n ")
    nvvidconv1 = Gst.ElementFactory.make("nvvideoconvert", "convertor1")
    if not nvvidconv1:
        sys.stderr.write(" Unable to create nvvidconv1 \n")
    print("Creating filter1 \n ")
    caps1 = Gst.Caps.from_string("video/x-raw(memory:NVMM), format=RGBA")
    filter1 = Gst.ElementFactory.make("capsfilter", "filter1")
    if not filter1:
        sys.stderr.write(" Unable to get the caps filter1 \n")
    filter1.set_property("caps", caps1)
    print("Creating tiler \n ")
    tiler=Gst.ElementFactory.make("nvmultistreamtiler", "nvtiler")
    if not tiler:
        sys.stderr.write(" Unable to create tiler \n")
    print("Creating nvvidconv \n ")
    nvvidconv = Gst.ElementFactory.make("nvvideoconvert", "convertor")
    if not nvvidconv:
        sys.stderr.write(" Unable to create nvvidconv \n")
    print("Creating nvosd \n ")
    nvosd = Gst.ElementFactory.make("nvdsosd", "onscreendisplay")
    if not nvosd:
        sys.stderr.write(" Unable to create nvosd \n")
    if(is_aarch64()):
        print("Creating transform \n ")
        transform=Gst.ElementFactory.make("nvegltransform", "nvegl-transform")
        if not transform:
            sys.stderr.write(" Unable to create transform \n")

    print("Creating EGLSink \n")
    sink = Gst.ElementFactory.make("nveglglessink", "nvvideo-renderer")
    if not sink:
        sys.stderr.write(" Unable to create egl sink \n")

    if is_live:
        print("Atleast one of the sources is live")
        streammux.set_property('live-source', 1)

    streammux.set_property('width', 1920)
    streammux.set_property('height', 1080)
    streammux.set_property('batch-size', number_sources)
    streammux.set_property('batched-push-timeout', 1/30)
    pgie.set_property('config-file-path', "dstest_imagedata_config.txt")
    #Set properties of pgie and sgie
    sgie1.set_property('config-file-path', "dstest2_sgie1_config.txt")
    
    pgie_batch_size=pgie.get_property("batch-size")
    if(pgie_batch_size != number_sources):
        print("WARNING: Overriding infer-config batch-size",pgie_batch_size," with number of sources ", number_sources," \n")
        pgie.set_property("batch-size",number_sources)
    tiler_rows=int(math.sqrt(number_sources))
    tiler_columns=int(math.ceil((1.0*number_sources)/tiler_rows))
    tiler.set_property("rows",tiler_rows)
    tiler.set_property("columns",tiler_columns)
    tiler.set_property("width", TILED_OUTPUT_WIDTH)
    tiler.set_property("height", TILED_OUTPUT_HEIGHT)

    sink.set_property("sync", 0)

    if not is_aarch64():
        # Use CUDA unified memory in the pipeline so frames
        # can be easily accessed on CPU in Python.
        mem_type = int(pyds.NVBUF_MEM_CUDA_UNIFIED)
        streammux.set_property("nvbuf-memory-type", mem_type)
        nvvidconv.set_property("nvbuf-memory-type", mem_type)
        nvvidconv1.set_property("nvbuf-memory-type", mem_type)
        tiler.set_property("nvbuf-memory-type", mem_type)
    config = configparser.ConfigParser()
    config.read('dstest2_tracker_config.txt')
    config.sections()
    for key in config['tracker']:
        if key == 'tracker-width':
            tracker_width = config.getint('tracker', key)
            tracker.set_property('tracker-width', tracker_width)
        if key == 'tracker-height' :
            tracker_height = config.getint('tracker', key)
            tracker.set_property('tracker-height', tracker_height)
        if key == 'gpu-id' :
            tracker_gpu_id = config.getint('tracker', key)
            tracker.set_property('gpu_id', tracker_gpu_id)
        if key == 'll-lib-file' :
            tracker_ll_lib_file = config.get('tracker', key)
            tracker.set_property('ll-lib-file', tracker_ll_lib_file)
        if key == 'll-config-file' :
            tracker_ll_config_file = config.get('tracker', key)
            tracker.set_property('ll-config-file', tracker_ll_config_file)
        if key == 'enable-batch-process' :
            tracker_enable_batch_process = config.getint('tracker', key)
            tracker.set_property('enable_batch_process', tracker_enable_batch_process)
        
    print("Adding elements to Pipeline \n")
    self.pipeline.add(pgie)
    self.pipeline.add(tracker)
    #self.pipeline.add(sgie1)
    self.pipeline.add(tiler)
    self.pipeline.add(nvvidconv)
    self.pipeline.add(filter1)
    self.pipeline.add(nvvidconv1)
    self.pipeline.add(nvosd)
    if is_aarch64():
        self.pipeline.add(transform)
    self.pipeline.add(sink)

    print("Linking elements in the Pipeline \n")
    streammux.link(pgie)  
    pgie.link(tracker)
    tracker.link(nvvidconv1)
    #pgie.link(nvvidconv1)
    nvvidconv1.link(filter1)
    filter1.link(tiler)
    tiler.link(nvvidconv)
    nvvidconv.link(nvosd)
    if is_aarch64():
        nvosd.link(transform)
        transform.link(sink)
    else:
        nvosd.link(sink)

    # create an event loop and feed gstreamer bus mesages to it
    loop = GObject.MainLoop()
    bus = self.pipeline.get_bus()
    bus.add_signal_watch()
    bus.connect ("message", bus_call, loop)
    tiler_sink_pad=tiler.get_static_pad("sink")
    if not tiler_sink_pad:
        sys.stderr.write(" Unable to get src pad \n")
    else:

        tiler_sink_pad.add_probe(Gst.PadProbeType.BUFFER, self.tiler_sink_pad_buffer_probe, 0)
    # List the sources
    print("Now playing...")
    for i, source in enumerate(args[:-1]):
        if (i != 0):
            print(i, ": ", source)
    print("Starting pipeline \n")
    # start play back and listed to events		
    self.pipeline.set_state(Gst.State.PLAYING)
    
    try:
        loop.run()
    except:
        pass
    # cleanup
    print("Exiting app\n")
    
    self.pipeline.set_state(Gst.State.NULL)

Hi,
cv2.resize() may take certain CPU usage. Please run sudo tegraststs to check the system loading.
For resizing, you may consider to use nvvideoconvert to leverage hardware converter.

Jtop :


So it seems i still have plenty of GPU and CPU to utilize and even without resize i’m still getting delay issues.
Can you explain in detail how to incorporate changes in resolution with nvvideoconvert specifically in the python example i use?

I’ve also seen solutions to do with queue leaky or num-extra-surfaces as seen here:

https://docs.nvidia.com/metropolis/deepstream/dev-guide/index.html#page/DeepStream%20Plugins%20Development%20Guide/deepstream_plugin_troubleshooting.html#

But how do i implement them in my code? or a is it in a different file? if so, which file?

Also, can you suggest a way to reset the stream on the fly? One which is different than the aforementioned method i tried.

Thanks.

Hi,
We have public python samples in

Please check and share which is close to your usecase. We would need to reproduce the issue first and do checking. See if we can have further optimization.

Resizing the image can be done through NvBufSurface APIs in C. We are still working on the python bindings for this.

There are several topics about pythons bindings going on:

Hi, thanks for your reply.
I appreciate you sharing the python samples link but as mentioned in the first post and seen in the attached code, i’m using the python sample “deepstream-imagedata-multistream”.
I’ve scoured through the different posts on this forum and have yet to find a solution. Any tips on how to reset the stream?

Hi,
If you run deepstream-imagedata-multistream,please configure nveglglessink sync=false and try again. It is to disable synchronization mechanism in gstreamer and might help.

As seen in the code i posted, sync was already set to “0” which i know to be “false”.
But, i tried changing to false anyway and there was no improvement.

Hi,
A optimal solution of resizing the image it to call NvBufferTransform() to leverage hardware converter. However, the API is only in C and has not been implemented in python bindings.

As of now we don’t have a better solution for this usecase.

Hi,
In the python code, it looks like self.dim1, self.dim2, self.dim3 are different so you have to call cv2.resize() three times. If the setting is identical, you may call cv2.resize() once and call np.copy() to duplicate the images.

For your reference. In C code, you can check gst-dsexample in

/opt/nvidia/deepstream/deepstream-5.0/sources/gst-plugins/gst-dsexample/

There is code of demonstrating NvBufSurf APIs.

Hi,
tried that, unfortunately the only way i can get rid of the delay is resizing only 1 of them to 720X480 (or lower) and setting only 1 of the Qlabels, whilst canceling all other operations. If i add in any additional operations the delay starts building up.

Can you advise me on how to reset the rtsp stream on the fly?