trt_batched_nms

ROS 2 package for TensorRT batched NMS plugin developed by NVIDIA and modified by mmdeploy.

Original source code:

• https://github.com/NVIDIA/TensorRT/tree/release/10.3/plugin/batchedNMSPlugin
• https://github.com/open-mmlab/mmdeploy/tree/main/csrc/mmdeploy/backend_ops/tensorrt/batched_nms

Compile the package:

colcon build --symlink-install --cmake-args -DCMAKE_BUILD_TYPE=Release

Usage

After compiling the package, shared library libtrt_batched_nms_plugin.so will be generated in the install/trt_batched_nms/lib directory.

You can use following logic to use generated shared library in cpp code:

const std::string libPath = "/path/to/libtrt_batched_nms_plugin.so";
void *handle = dlopen(libPath, RTLD_LAZY);

Also, you can use following example in your launch or config files to use generated shared library as parameter in your ROS 2 packages.

• "$(find-pkg-prefix trt_batched_nms)/lib/libtrt_batched_nms_plugin.so"