TensorRT Optimizations

[Ishaan-Datta]

Inference:

• [ ] Non-maximum supression (NMS)
• [ ] NMS Example
• [ ] Example for end-to-end YOLO-ONNX TRT
• [ ] Another Example for ONNX YOLO TRT
• [ ] Example including batching

Zed:

• [x] Generic C++ Application bulding
• [x] OpenCV Integration
• [x] Assigning GPU to zed camera

Nvidia Video Tutorials:

• [ ] Jetson AI Fundamentals - S3E2 - Image Classification Inference - YouTube
• [ ] Jetson AI Fundamentals - S3E4 - Object Detection Inference - YouTube
• [ ] Jetson AI Fundamentals - S3E5 - Training Object Detection Models - YouTube
• [ ] Jetson AI Fundamentals - S3E6 - Semantic Segmentation - YouTube
• [ ] Inference Optimization with NVIDIA TensorRT - YouTube
• [ ] Boosting Performance and Utilization with Multi-Instance GPU - YouTube
• [ ] Running Multiple Workloads on a Single A100 GPU - YouTube

Asynchronous Processing:

• [ ] Try offloading any preprocessing and postprocessing to dedicated nodes and measure any performance benefits (latency)
• [ ] Investigate how multithreaded ROS2 Nodes could fit into our architecture and measure any performance benefits

gstreamer