Trying to access frame using "pyds.get_nvds_buf_surface(hash(gst_buffer), frame_meta.batch_id)" but instead encountering with Segmentation fault

Accelerated Computing Intelligent Video Analytics DeepStream SDK
#1

Hi I am working on python application test 1. I want to save bounding box region in memory. I am trying to access frame using below code inside
def osd_sink_pad_buffer_probe(pad,info,u_data):

n_frame=pyds.get_nvds_buf_surface(hash(gst_buffer), frame_meta.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)

But I am facing segmentation fault error after processing of few frames. The same configuration and app file is working fine if I am not accesing frames.

• Hardware Platform (GPU)–> Tesla k40m
• DeepStream Version 5.0
• NVIDIA GPU Driver Version (valid for GPU only) -->440.100
• Issue Type( bugs)

#3

Hey, we have a demo https://github.com/NVIDIA-AI-IOT/deepstream_python_apps/blob/5cb4cb8be92e079acd07d911d265946580ea81cd/apps/deepstream-imagedata-multistream/deepstream_imagedata-multistream.py , have you checked it?

#4

Hi , bcao. I have worked on this demo. But it works for me for few frames then stops, showing “bus error”. On changing some conditions the pipeline stops at starting. The code I have mentioned earlier was taken from this example only.

#5

So the demo can work for you, right?
If yes, can you share all your code for debug?

#7

Yes it is working , but in unreliable way, sometimes it runs to thousands of frames, sometimes just in hundred and stops abruptly.
something like this…

Screenshot (303)
Screenshot (303)1484×837 78.5 KB

I am using rtsp stream.

My script is–
test_image_deepstream.py (16.0 KB)

How to take photo (How to save frame image)
#8

I think there is problem while integration of Opencv with gst plugin in my pipeline. I have added below in deepstream reference app after pgie , this is also generating bus error for me
[ds-example]
enable=1
processing-width=640
processing-height=480
full-frame=0
unique-id=15

#9

Hi, I have found this stament in https://docs.nvidia.com/metropolis/deepstream/DeepStream_5.0_Release_Notes.pdf -
“On NVIDIA Tesla T4 with driver 450.51, using NVENC sometimes results into a CUDA
error. An alternative is to use software encoder for file and RTSP output.”

And I am using Tesla k40m. is this the reason for me getting bus error???

#10

Sry for the delay,

Just to confirm, so the original demo which doesn’t include your own modification also stops abruptly and prints a bus error, right?

#11

Yes it ends, when it tries to create the buffer to access the frame
i.e when the conditions satifies to run –
n_frame=pyds.get_nvds_buf_surface(hash(gst_buffer),frame_meta.batch_id).

Sometimes it saves few frames with bboxes, but eventually stops with bus error.

#12

Any reason you need to use the old nvidia driver, we suggest to use:

You must install the following components: - Ubuntu 18.04 - GStreamer 1.14.1 - NVIDIA driver 450.51 - CUDA 10.2 - TensorRT 7.0.X

https://docs.nvidia.com/metropolis/deepstream/dev-guide/text/DS_Quickstart.html

#14

Dear sir, the above segmentation fault is still showing on my system, I have updated my deepstream 5.0 to 5.1, I am using driver 460.32, Cuda 11.1, My Gpus are Tesla K80, gstreamer 1.14.5,Tensort 7.2.2 , ubuntu 18.04. In my application extracting frame meta is important task, And I am stuck with it. Please provide me a solution, if not possible atleast explain me possible reasons for it so that I could work on it. All other functionality is working except when I try to extract frames using –
n_frame = pyds.get_nvds_buf_surface(hash(gst_buffer), frame_meta.batch_id)

#16

@bhatiyaarpit95 have you solved the problem, does the original deepstream_python_apps/apps/deepstream-imagedata-multistream at master · NVIDIA-AI-IOT/deepstream_pyt work?

#17

Hi bcao, It is still unresolved. My original imagedata- code is working until it meets a condition where it saves images then there is bus error. Maybe our application is not getting the pointer or memory access, i don’t know…

#18

Also @bcao I also have a question regarding custom classification models and object detection models other than provided in samples. I am a python developer so its hard for me to figure out, if possible please help me with custom implementations…

#21

But i can not repro segmentation fault issue or bus error, i use original deepstream-imagedata-multistream and builtin stream sample_720p.h264, is there any difference between us?

Frame Number= 1433 Number of Objects= 8 Vehicle_count= 6 Person_count= 2
Frame Number= 1434 Number of Objects= 9 Vehicle_count= 7 Person_count= 2
Frame Number= 1435 Number of Objects= 9 Vehicle_count= 7 Person_count= 2
Frame Number= 1436 Number of Objects= 9 Vehicle_count= 7 Person_count= 2
Frame Number= 1437 Number of Objects= 9 Vehicle_count= 7 Person_count= 2
Frame Number= 1438 Number of Objects= 10 Vehicle_count= 8 Person_count= 2
Frame Number= 1439 Number of Objects= 8 Vehicle_count= 6 Person_count= 2
Frame Number= 1440 Number of Objects= 8 Vehicle_count= 6 Person_count= 2
Frame Number= 1441 Number of Objects= 0 Vehicle_count= 0 Person_count= 0
End-of-stream
Exiting app