Object Detection with TensorRT


I have been trying to run Object detection with Tensorrt. I first noticed on the github page for detectnet (GitHub - NVIDIA-ISAAC-ROS/isaac_ros_object_detection: Deep learning model support for object detection including DetectNet) that it says:

Note: The Isaac ROS TensorRT package is not able to perform inference with DetectNet models at this time.

What is the exact reason for this statement?

Furthermore I tried running FasterRCNN in Tensorrt and I got:

[executionContext.cpp::enqueueV2::314] Error Code 3: API Usage Error (Parameter check failed at: runtime/api/executionContext.cpp::enqueueV2::314, condition: !mEngine.hasImplicitBatchDimension()

I converted my .etlt that was trained in TLT 3.0 to a .plan. Do you have any idea what the error is?

Best wishes

The Isaac ROS TensorRT package does not yet support configuring TensorRT for implicit batch dimensions. However, Isaac ROS Triton with the TensorRT backend does support implicit batch dimensions correctly, and should get you similar performance.