Pipeline paused after Color primaries 5 not present and will be treated BT.601

[nam1012.2016]

Please provide complete information as applicable to your setup.
• Hardware Platform: dGPU
**• DeepStream Version: 7.0 **
• TensorRT Version: 8.6.1
• NVIDIA GPU Driver Version: 550
• Issue Type: bugs
My pipeline stopped when one of my source connect and get warning Color primaries 5 not present and will be treated BT.601 . It look something like this. If

**PERF:  {'stream0': 0.0, 'stream1': 0.0, 'stream2': 0.0, 'stream3': 20.41} 

Warning: Color primaries 5 not present and will be treated BT.601

**PERF:  {'stream0': 4.05, 'stream1': 0.0, 'stream2': 0.0, 'stream3': 8.2} 


**PERF:  {'stream0': 0.0, 'stream1': 0.0, 'stream2': 0.0, 'stream3': 0.0} 


**PERF:  {'stream0': 0.0, 'stream1': 0.0, 'stream2': 0.0, 'stream3': 0.0}

Here is my log and pipeline graph:
gs.log (1.5 MB)

output.dot.txt (214.1 KB)
The state change started from streaming_bin_3. I dont know what caused this, maybe the h264 source is corrupted but i dont want my whole pipeline to go down if one source is corrupted.

[nam1012.2016]

Well it turns out i have to put nvvideoconvert after demux and set disable-passthrough to True.

[junshengy]

Glad to hear that, noticed your pipeline uses tee/metamux which seems unnecessary，

                        |  --> nvinfer --> nvinfer ---> |                      
nvstreammux --> tee -->                                 | --> metamux --> tee --> 
                        | --> nvinfer ----------------> |  

If you want parallel inference, please refer to this topic

NvdsMetaMux not merging metadata

I think this is normal. The source id of each nvstreammux is independent, which means that the source ids of the two nvstreammux in the above pipeline are both 0/1, which cannot be distinguished when merging. So if you want to merge metadata from multiple branches, you can only refer to the pipeline of the parallel sample app. First, use nvstreammux to form all streams into a batch, then nvstreamdemux, nvstreammux, and finally merge them using metamux. [image]

[nam1012.2016]

Thanks for your suggestion. The tiler branch is for multistream visualize. while the others branch after demuxer is for full frame image saving with bounding box. So
with your suggestion, what should i do to replace tee after metamux? Should i just do

                         |nvvideoconvert -> osd |
metamux --> demuxer -->                          --> streammux -> tiler    
                         |nvvideoconvert -> osd |

instead?

[junshengy]

The following pipeline should be similar to your needs. Add queue or nvvvideoconvert where needed.

                                                                             | --> tiler
nvstreammux -->  nviner (pgie1 )--> nvinfer (pgie 2)--> infer(sgie2) --> tee | 
                                                                             | --> nvstreamdemux --> |  --> source 1
                                                                                                     |  --> source 2

[nam1012.2016]

Absolutely love it, thanks you so much.