Nvvideoconvert sabotages the NvDsBatchMeta

[ahmed.t.mohamed]

Hi,
I am running Deepstream 5.1 on Dell G5 15 the GPU is GTX 1650ti the OS is Ubuntu 18.04.5LTS tensorrt version 7.2.3.4 GPU Driver 460.56 CUDA version 11.2

I have this pipeline decodebin->nvstreammux->nvinfer->nvstreamtiler->nvvideoconvert->nvosd->tee->nvmsgconv->nvmsgbroker
|
|
->nveglglessink

The NvDsBatchMeta before the nvvideoconvert have the right batch size for example if I start the streaming with 2 sources it have batch size 2 in num_frames_in_batch but after the nvvideoconvert it only returns batch size 1 in the NvDsBatchMeta and the buffer also have only 1 image from the batch when I extract the surface buffer from the buffer.

I am not setting any options for nvvideoconvert in my pipeline.

[Fiona.Chen]

nvstreamtiler will combine the multiple streams into only one stream. So the batch size is always 1 after nvstreamtiler.

[ahmed.t.mohamed]

When I put the nvstreamtiler after the nvdsosd the bboxes is flashing and appears in wrong places.

[Fiona.Chen]

nvdsosd must be put after nvmultistreamtiler.

[ahmed.t.mohamed]

Thanks that was helpful. I have solved it now.