Hi,
A possible solution for this usecase is to run pipeline like:
... ! tee name=t ! queue ! nvvidconv ! fakesink t. ! queue ! appsink
In appsink plugin, you can call NvBufferTransform() to execute scaling with different setting for each frame. For accessing NvBuffer in appsink, please refer to
How to run RTP Camera in deepstream on Nano - #29 by DaneLLL
NvBuffer APIs are defined in
/usr/src/jetson_multimedia_api/include/nvbuf_utils.h