Segmentation fault core dumped, deepstream 6.2

[Taco]

• Hardware Platform (t4 GPU)
• DeepStream Version 6.2

• Ubuntu 20.04
• GStreamer 1.16.3
• NVIDIA driver 525.60.13
• CUDA 11.8
• TensorRT 8.5.1.7
  so basically im trying to access a frame from a video to write it and i ran it to a segmentation fault

def get_frame_from_gst_buffer(gst_buffer, batch_id,obj_meta):
    print(f"Entering get_frame_from_gst_buffer with batch_id: {batch_id}")
    
    if gst_buffer is None:
        print("gst_buffer is None")
        return None
    
    try:
        print(f"Calling get_nvds_buf_surface with hash(gst_buffer): {hash(gst_buffer)}")
        nvbufsurface = pyds.get_nvds_buf_surface(hash(gst_buffer), batch_id)
        print(f"nvbufsurface type: {type(nvbufsurface)}")
        
        if nvbufsurface is None:
            print("get_nvds_buf_surface returned None")
            return None
        
        print(f"nvbufsurface shape: {nvbufsurface.shape}, dtype: {nvbufsurface.dtype}")
        print(f"nvbufsurface memory info: {nvbufsurface.base}")
        
    except Exception as e:
        print(f"Exception in get_nvds_buf_surface: {e}")
        return None

    try:
        print("Attempting to create frame_copy")
        frame_copy = np.asarray(nvbufsurface, order='C')
        print(f"frame_copy shape: {frame_copy.shape}")
    except Exception as e:
        print(f"Exception in np.array: {e}")
        return None

    try:
        print("Attempting cv2.imencode")
        _, buffer = cv2.imencode('.jpg', frame_copy)
        print(f"Encoded buffer size: {len(buffer)}")
    except Exception as e:
        print(f"Exception in cv2.imencode: {e}")
        return None

    image_base64 = base64.b64encode(buffer).decode()
    print(f"base64 encoded image length: {len(image_base64)}")

    return image_base64

nvbufsurface type: <class ‘numpy.ndarray’>
nvbufsurface shape: (1080, 1920, 4), dtype: uint8
nvbufsurface memory info: <capsule object NULL at 0x7f1d09fd0960>
Attempting to create frame_copy
frame_copy shape: (1080, 1920, 4)
Attempting cv2.imencode
Segmentation fault (core dumped)
, can someone let me know why this is happening?

[junshengy]

1. Only RGBA format can be supported.

2. Use CUDA unified memory in the pipeline so frames can be easily accessed on CPU in Python.

refer this sample deepstream_python_apps/apps/deepstream-imagedata-multistream/deepstream_imagedata-multistream.py at v1.1.6 · NVIDIA-AI-IOT/deepstream_python_apps · GitHub

[Taco]

Alright so i added

    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)

    mem_type = int(pyds.NVBUF_MEM_CUDA_UNIFIED)
    streammux.set_property("nvbuf-memory-type", mem_type)

        nvvideoconvert.set_property("nvbuf-memory-type", mem_type)
        nvvideoconvert2.set_property("nvbuf-memory-type", mem_type)
  
        # connect  queue -> nvvidconv -> nvosd -> nveglgl
        queue.link(nvvideoconvert)
        nvvideoconvert.link(filter1)
        filter1.link(nvdsosd)
        nvdsosd.link(nvvideoconvert2)
        nvvideoconvert2.link(capsfilter)
        capsfilter.link(encoder)
        encoder.link(codeparser)
        if args[i].split(".").pop() == "h264":
            codeparser.link(sink)
        else:
            codeparser.link(container)
            container.link(sink)

        sink.set_property("qos", 0)

like from the example app , now im facing this error
Error: gst-stream-error-quark: memory type configured and i/p buffer mismatch ip_surf 2 muxer 3 (1): gstnvstreammux.c(643): gst_nvstreammux_chain (): /GstPipeline:pipeline0/GstNvStreamMux:Stream-muxer
i’ll attach my entire code
DeepstreamError.txt (22.9 KB)

[junshengy]

There are some issues with your pipeline, build the pipeline like below.

Is it mandatory to use mpeg4 encoding? Use hardware encoding to improve performance.

uridecodebin  --> nvstreammux --> gie -->nvvideoconvert --> capsfiller(NVMM:RGBA) --> nvosd --> nvsteamdemux --> nvvideoconvert --> cpafiller2(NVMM:NV12) --> nvv4l2h264enc --> mp4mux --> filesink
                                                         |
                                                 add probe for save_frame_image_as_base64

[Taco]

@junshengy

mp4mux

ah i’ll make those necessary changes for improved performance, but the above error was because i forgot to set object to cuda unified memory in the decodebin

def decodebin_child_added(child_proxy, Object, name, user_data):

    print("Decodebin child added:", name, "\n")
    if name.find("decodebin") != -1:
        Object.connect("child-added", decodebin_child_added, user_data)
    if name.find("nvv4l2decoder") != -1:
 
        Object.set_property("cudadec-memtype", 2)

    if "source" in name:
        source_element = child_proxy.get_by_name("source")
        if source_element.find_property("drop-on-latency") != None:
            Object.set_property("drop-on-latency", True)

Thank you for your assistance!