DGPU: T4
Docker: nvcr.io/nvidia/deepstream:5.0.1-20.09-devel
BUG
Hi Nvidia Team,
We are experiencing a blocking issue when upgrading our pipeline from DeepStream 4.0.2 to DeepStream 5.0.1: nvstreamdemux no longer works when using RGBA conversion, it produces a core dump.
RGBA conversion is mandatory when modifying image buffers like blurring before demuxing.
Some simple test cases to reproduce:
#1 Pipeline with RGBA conversion which produces a core dump:
nvstreammux nvbuf-memory-type=3 name=m batched-push-timeout=40000 batch-size=2 width=960 height=540 ! queue ! nvinfer batch-size=2 config-file-path=/path_to_model/trafficCamNet/config_infer_primary.txt ! queue ! nvvideoconvert nvbuf-memory-type=3 ! “video/x-raw(memory:NVMM), format=RGBA” ! nvstreamdemux name=demux demux.src_0 ! nvvideoconvert nvbuf-memory-type=3 ! nvdsosd ! nveglglessink qos= true sync=false async=false demux.src_1 ! nvvideoconvert nvbuf-memory-type=3 ! nvdsosd ! nveglglessink qos= true sync=false async=false uridecodebin uri=file:///opt/nvidia/deepstream/deepstream-5.0/samples/streams/sample_1080p_h265.mp4 name=decodebin_0 ! m.sink_0 uridecodebin uri=file:///opt/nvidia/deepstream/deepstream-5.0/samples/streams/sample_1080p_h265.mp4 name=decodebin_1 ! m.sink_1
#2 Same pipeline with NV12 which works properly :
nvstreammux nvbuf-memory-type=3 name=m batched-push-timeout=40000 batch-size=2 width=960 height=540 ! queue ! nvinfer batch-size=2 config-file-path=/path_to_model/trafficCamNet/config_infer_primary.txt ! queue ! nvvideoconvert nvbuf-memory-type=3 ! “video/x-raw(memory:NVMM), format=NV12” ! nvstreamdemux name=demux demux.src_0 ! nvvideoconvert nvbuf-memory-type=3 ! nvdsosd ! nveglglessink qos= true sync=false async=false demux.src_1 ! nvvideoconvert nvbuf-memory-type=3 ! nvdsosd ! nveglglessink qos= true sync=false async=false uridecodebin uri=file:///opt/nvidia/deepstream/deepstream-5.0/samples/streams/sample_1080p_h265.mp4 name=decodebin_0 ! m.sink_0 uridecodebin uri=file:///opt/nvidia/deepstream/deepstream-5.0/samples/streams/sample_1080p_h265.mp4 name=decodebin_1 ! m.sink_1
3# Pipeline with double conversion NV12 → RGBA → NV12 witch also produces a core dump :
nvstreammux nvbuf-memory-type=3 name=m batched-push-timeout=40000 batch-size=2 width=960 height=540 ! queue ! nvinfer batch-size=2 config-file-path=/path_to_model/trafficCamNet/config_infer_primary.txt ! queue ! nvvideoconvert nvbuf-memory-type=3 ! “video/x-raw(memory:NVMM), format=RGBA” ! queue ! nvvideoconvert nvbuf-memory-type=3 ! “video/x-raw(memory:NVMM), format=NV12” ! nvstreamdemux name=demux demux.src_0 ! nvvideoconvert nvbuf-memory-type=3 ! nvdsosd ! nveglglessink qos= true sync=false async=false demux.src_1 ! nvvideoconvert nvbuf-memory-type=3 ! nvdsosd ! nveglglessink qos= true sync=false async=false uridecodebin uri=file:///opt/nvidia/deepstream/deepstream-5.0/samples/streams/sample_1080p_h265.mp4 name=decodebin_0 ! m.sink_0 uridecodebin uri=file:///opt/nvidia/deepstream/deepstream-5.0/samples/streams/sample_1080p_h265.mp4 name=decodebin_1 ! m.sink_1
All these tests were running perfectly with nvcr.io/nvidia/deepstream:4.0.2-19.12-devel
Please address this issue as a major one.