Hi, I would like to divide one frame into mutiple images with smaller size and pack them as a batch to get the infer results using deepstream nvinfer plugin. Next, I would like to resume the orignal frame in display and show the bbox results on it . Could you give me some solution or advise about it?
Our nvvideoconvert supports crop function , it may help you to split the video/image to different parts. Gst-nvvideoconvert — DeepStream 5.1 Release documentation
But we don’t have plugin to combine the image back to the original one, you may try to use nvmultistreamtiler to combine the divided videos together.
The following pipeline is a sample:
gst-launch-1.0 uridecodebin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h264.mp4 ! nvvideoconvert src-crop=0:0:960:1080 ! mux.sink_0 nvstreammux name=mux width=1920 height=1080 batch-size=2 ! nvmultistreamtiler rows=1 columns=2 width=1920 height=1080 ! nvegltransform ! nveglglessink uridecodebin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h264.mp4 ! nvvideoconvert src-crop=960:0:960:1080 ! mux.sink_1
Great demo! I have checked nvmultistreamtiler plugin document and found custom-tile-config is still under development. If I want to divide one frame into different size with some overlap to aviod the effect on the detection results. The combined frame after nvmultistreamtiler will be strange. How to solve that?
No. Current deepstream can not handle such case.
This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.