Dynamic boxes from PGIE object detection and cropping (precess mode=2) to SGIE action recognition

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Description

I am trying to run the Action Recognition example with the Region of Interest (ROI) being the inference result from the Primary GIE (PGIE). In my example, I am receiving bounding boxes from the PeopleNet PGIE, which are intended to serve as the ROI for the Action Recognition application in DeepStream examples.

My simplified pipeline looks like this:
filesource > nvstreammux > nvinfer > nvtracker > nvdspreprocess > nvinfer > eglsink

Models:

  1. I have used the peoplenet model from Nvidia for PGIE (https://api.ngc.nvidia.com/v2/models/nvidia/tao/peoplenet/versions/deployable_quantized_v2.6.1/files)
  2. I have used default Action Recognition model from Nvidia for SGIE (https://api.ngc.nvidia.com/v2/models/nvidia/tao/actionrecognitionnet/versions/deployable_v1.0/files) with classes push, walk, run etc.

I have refered the example : Action Recognition on Dynamic ROIs fails with CUDA error

I followed the steps provided in the link above, but I am encountering a segmentation fault when running the pipeline.

Difference wrt above example is :

  1. The referred example was tried on deepstream 6.2, and the moderator indicated that the issue might have been fixed with 6.3. I am using deepstream 7.0
  2. Minor config file differences

The segmentation fault output on the terminal is below:

Decodebin child added: nvv4l2decoder0
In cb_newpad
[DEBUG: DS_3DAR] Infer done for frame 0, number of objects: 0
[tanmay:115 :0:137] Caught signal 11 (Segmentation fault: address not mapped to object at address 0x18)
==== backtrace (tid: 137) ====
0 0x0000000000042520 __sigaction() ???:0
1 0x000000000001c4a2 NvTrackerProc::processBatch() ???:0
2 0x00000000000dc253 std::error_code::default_error_condition() ???:0
3 0x0000000000094ac3 pthread_condattr_setpshared() ???:0
4 0x0000000000125a04 clone() ???:0

Segmentation fault (core dumped)

The code and config file that I have used for setting this up is attached below.

Could you please guide me on any additional modifications or troubleshooting steps I should consider to make the Action Recognition run smoothly as a secondary GIE with dynamic boxes from PGIE?

Thank you in advance for your assistance.

Environment

TensorRT Version:8.5.2.2
GPU Type: NVIDIA GeForce RTX 3070 Ti 8GB VRAM
Nvidia Driver Version: 535
DeepStream Version: 7.0 (Docker)

Relevant Files

3

I think you can refer to this sample.

4

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

5

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