Segmetations fault occurs when running deepstream7.0 docker with wsl2

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

• Deep Stream Version

  • 7.0
    • JetPack version (valid only for Jetson)
    • TensorRT version
  • 8.6.1
    • NVIDIA GPU driver version (valid for GPU only)
  • 552.22
    • Issue Type (Question, New Requirement, Bug)
  • Segmentation fault occurs when running deepstream7.0 docker using wsl2.
  • If you apply the following syntax in the process of saving the image in frames after object detection, an error called segmentations fault occurs.
  • Please note that if you delete this phrase, the image will not be saved, but the fps will be normal and the segmentations fault error will not occur.
  • n_frame = pyds.get_nvds_buf_surface(hash(gst_buffer), frame_meta.batch_id)
    frame_copy = np.array(n_frame, copy=True, order=‘C’)
    frame_copy = cv2.cvtColor(frame_copy, cv2.COLOR_RGBA2BGRA)
    json_data = self.__make_json(cctv_id,object_list)
    cv2.imwrite(img_path, frame_copy)

• How do I reproduce the problem? (This is for a bug. Include which sample app you are using, the contents of the configuration file, the command line used, and other details for reproduction.)
• Requirement details (this is for a new requirement, including the module name - which plugin it is for, which sample application it is for, a description of its functionality)

3

Please refer to this example. I have tested it on WSL and it works. However, since DeepStream on WSL2 is an alpha release, there are some problems with the OSD output.

/opt/nvidia/deepstream/deepstream/sources/deepstream_python_apps/apps/deepstream-imagedata-multistream

Please modify the code according to the following example

  if not platform_info.is_integrated_gpu():
        # 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)
        if platform_info.is_wsl():
            #opencv functions like cv2.line and cv2.putText is not able to access NVBUF_MEM_CUDA_UNIFIED memory
            #in WSL systems due to some reason and gives SEGFAULT. Use NVBUF_MEM_CUDA_PINNED memory for such
            #usecases in WSL. Here, nvvidconv1's buffer is used in tiler sink pad probe and cv2 operations are
            #done on that.
            print("using nvbuf_mem_cuda_pinned memory for nvvidconv1\n")
            vc_mem_type = int(pyds.NVBUF_MEM_CUDA_PINNED)
            nvvidconv1.set_property("nvbuf-memory-type", vc_mem_type)
        else:
            nvvidconv1.set_property("nvbuf-memory-type", mem_type)
        tiler.set_property("nvbuf-memory-type", mem_type)
4

If you run the file in /opt/nvidia/deepstream/deepstream/sources/deepstream_python_apps/apps/deepstream-imagedata-multistream, the image will also be saved.
However, there is no difference between that method and my image saving method.
However, a segmentations fault error occurs.
I don’t know why.
wsl so vc_mem_type = int(pyds.NVBUF_MEM_CUDA_PINNED)
I applied nvvidconv1.set_property(“nvbuf-memory-type”, vc_mem_type), but the same problem occurs.
The problematic syntax is n_frame = pyds.get_nvds_buf_surface(hash(gst_buffer), frame_meta.batch_id)
This is part.
I confirmed that values ​​such as hash(gst_buffer) are output as 140705477663088 and frame_meta.batch_id as 0, but the above syntax does not seem to work.

5

Please share a sample that can reproduce the problem. I think this problem is caused by the difference between your pipeline and deepstream_imagedata-multistream.py.

In addition, as I said above, DS-7.0 on WSL is an alpha release, there will be some bugs. You can test it on native linux.

6

I found the cause.
This was a problem caused by applying mem_type = int(pyds.NVBUF_MEM_CUDA_UNIFIED) to nvstreammux. The problem occurred because mem_type = int(pyds.NVBUF_MEM_CUDA_PINNED) should have been applied here as well.
Thank you for your interest in the current issue.

7

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