支持多层输出的TensorRT

Face-Recognition =========================== This sample targets for demonstrating TensorRT2.1 Plugin API We leverage most of the functions from jetson_inference; please check it first if you need more DL samples: <https://github.com/dusty-nv/jetson-inference> *** </br> ## Environment Jetson TX2 </br> JetPack-3.1 </br> </br> </br> ## Build ```C $ sudo apt-get install git cmake $ git clone https://github.com/AastaNV/Face-Recognition $ cd Face-Recognition $ mkdir build $ cd build $ cmake .. $ make ``` </br> ## Run ```C $ cd aarch64/bin $ ./face-recognition ``` </br> ## Plugin Layer **BboxMergeLayer** </br> This plugin layer demonstrate how to implement a CPU-based Plugin layer </br> 1. Make required tensor as output 2. Allocate unified memory: CPU pointer== GPU pointer </br> **DataRoiLayer** </br> This plugin layer demonstrate how to implement a GPU Plugin layer </br> 1. Got input/output data pointer in enqueue function 2. Launch GPU kernel with same cuda stream </br> **RecognitionLayer** </br> This plugin layer demonstrate more complicated handling of Plugin layer </br> 1. This class can handle two differient layers: selectBbox and summaryLabel 2. Define some shared variable to make between layers communication easier </br> </br> ## Support Please rise your problem in our forum to get immediately support. </br> https://devtalk.nvidia.com/default/board/189/jetson-tx2/ </br> </br> </br>