Object Detection/Segmentation Skill

Overview

Expert skill for deploying and optimizing deep learning models for object detection, instance segmentation, and 3D object detection in robotics applications.

Capabilities

• Configure YOLO (v5, v8) for real-time detection
• Set up Detectron2 for instance segmentation
• Implement semantic segmentation models
• Configure TensorRT optimization for Jetson
• Set up ONNX runtime deployment
• Implement 3D object detection (PointPillars, VoxelNet)
• Configure depth-based object detection
• Set up ROS vision pipelines with image_pipeline
• Implement object tracking (SORT, DeepSORT, ByteTrack)
• Configure multi-camera detection fusion

Target Processes

• object-detection-pipeline.js
• synthetic-data-pipeline.js
• nn-model-optimization.js
• moveit-manipulation-planning.js

Dependencies

• YOLO (Ultralytics)
• Detectron2
• TensorRT
• ONNX Runtime
• vision_msgs

Usage Context

This skill is invoked when processes require object detection model deployment, instance segmentation, 3D detection, or multi-object tracking for robot perception.

Output Artifacts

• Detection model configurations
• TensorRT optimized models
• ROS detection node implementations
• Tracking pipeline configurations
• Multi-camera fusion setups
• Inference optimization scripts