Issue: RuntimeError: get_nvds_buf_Surface: Currently we only support RGBA color Format

Accelerated Computing Intelligent Video Analytics DeepStream SDK
,
1

hi,
I’m encountering an issue while working with DeepStream 6.3 and CUDA 12.2 in my pipeline. Specifically, I get the following error: Currently we only support RGBA color Format
To resolve this, I attempted to set the memory management to CUDA unified memory using the following code:

mem_type = int(pyds.NVBUF_MEM_CUDA_UNIFIED)
nvvidconv.set_property(“nvbuf-memory-type”, mem_type)
streammux.set_property(“nvbuf-memory-type”, mem_type)

However, I ran into another issue when applying this solution: TypeError: object of type GstNvStreamMux does not have property nvbuf-memory-type
Has anyone experienced a similar problem or knows how to properly set the memory type for the GstNvStreamMux and nvvidconv elements?

3

Are you using Jetson? You don’t need to set this properties on Jetson.

Just add nvvideoconvert to the upstream element which you add probe function.

Refer to /opt/nvidia/deepstream/deepstream/sources/deepstream_python_apps/apps/deepstream-imagedata-multistream/deepstream_imagedata-multistream.py

6

I’m working on an NVIDIA T4 server and I’m trying to save detected faces. I’ve added nvvideoconvert and *‘filter’

nvvideoconvert = Gst.ElementFactory.make(“nvvideoconvert”, “nvvideoconvert”)
if not nvvideoconvert:
sys.stderr.write(" Unable to create nvvideoconvert \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)

I also set the memory type like this:
if not platform_info.is_integrated_gpu():
mem_type = int(pyds.NVBUF_MEM_CUDA_UNIFIED)
streammux.set_property(“nvbuf-memory-type”, mem_type)
nvvideoconvert.set_property(“nvbuf-memory-type”, mem_type)
if platform_info.is_wsl():
print(“using nvbuf_mem_cuda_pinned memory for nvvidconv1\n”)
vc_mem_type = int(pyds.NVBUF_MEM_CUDA_PINNED)
nvvideoconvert.set_property(“nvbuf-memory-type”, vc_mem_type)
else:
nvvideoconvert.set_property(“nvbuf-memory-type”, mem_type)

here’s how I’m linking the elements in the pipeline :

streammux.link(queue1)
queue1.link(pgie)
pgie.link(tracker)
tracker.link(nvanalytics)
nvanalytics.link(pgie1)
pgie1.link(sgie2)
sgie2.link(sgie3)
sgie3.link(sgie4)
sgie4.link(nvvideoconvert)
nvvideoconvert.link(filter1)
filter1.link(nvstreamdemux)

However, I still encounter the same error.

7

Are you using the new nvstreammux? new nvstreammux cannot set this property.

You can get the configurable properties by running the following command:

USE_NEW_NVSTREAMMUX=yes gst-inspect-1.0 nvstreammux
8

Yes, I have switched to the old nvstreammux. The pipeline runs without any errors, but it still does not save the faces.

9

If you want to save the object as an image, you can refer to nvds_obj_enc_process.

This is sample code.
/opt/nvidia/deepstream/deepstream/sources/apps/sample_apps/deepstream-image-meta-test/deepstream_image_meta_test.c

However, there is no python bindinds at present, so you need to bind the C API to the Python code by yourself.

10

I added the corresponding python bindings and provided a sample code based on test1.

Refer this FAQ, thanks