Dynamic model swapping for nvinfer

With the nvinfer plugin for DeepStream, is it possible to dynamically swap the TensorRT engine at runtime, will swapping models require destroying the current pipeline and creating a new one, or is it possible to swap one nvinfer element with another as in this example?

DS 5.0 can support on-the-fly model update, please refer deepstream-tes5 sample and its README

1 Like