Deepstream process data in singular buffer manner without tiler

[fanzh]

1. seems nvvidconv is used in your pipeline. As written in this topic, please use nvvideoconvert instead of nvvidconv.
2. if there are three sources, after nvstreammux, the batch will includes three surfaces. nvjpegenc will encode one surface. but the relation of surface id and source id are not fixed, As mentioned in my first comment. Hence as you said, the images inside it are a mix of frames from many sources. Here are some resolutions.
   a> you can use nvstreamdemux to demux each surface, then add nvjpegenc.
   b> if inference is not needed, you can use three “source ->nvvideoconvert->nvjpegenc > fakesink” pipelines.