Nvinfer sometimes error when modifying the pipeline dynamically

Hi,
I use the nvcompositor plugin to composite two AI branches and then preview with nv3dsink. When try to add one branch (look like v4l2src → nvvideoconvert → … → nvcompositor → tee name=t → queue → nv3dsink t. → queue → … → filesink, the t. → queue → … → filesink branch is new one created dynamically), sometimes work well while sometimes nvinfer will output error:
ERROR: from element /GstPipeline:pipeline0/GstNvInfer:nvinfer2: NvBufSurfTransform failed with error -3 while converting buffer Additional debug info: gstnvinfer.cpp(1397): convert_batch_and_push_to_input_thread (): /GstPipeline:pipeline0/GstNvInfer:nvinfer2
And the output window is green.
Any help to correct this would be appreciated.

Hi,
In DeepStream SDK, please use nvmultistreamtiler plugin. The nvcompositor is not applicable in DeepStream SDK.

I have modified the nvcompositor to make it work in the static pipeline, while when I add one record branch after my nvcompositor plugin dynamically(block the tee’s sink pad, set the whole pipeline state PAUSED, add record branch and unblock), sometimes works but sometimes not. The error ofen occurs in the second AI branch’s nvinfer plugin:

ERROR: from element /GstPipeline:pipeline0/GstNvInfer:nvinfer2: NvBufSurfTransform failed with error -3 while converting buffer
Additional debug info: gstnvinfer.cpp(1397): convert_batch_and_push_to_input_thread ():

So the error is caused by the nvcompositor plugin ?

Hi,
This is not the user-case we have verified/tested with the default nvcompositor. It may not work properly.

For adding/deleting sources dynamically, please take a look at this sample:
deepstream_reference_apps/runtime_source_add_delete at master · NVIDIA-AI-IOT/deepstream_reference_apps · GitHub

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