senior-computer-vision

What This Skill Does

The senior-computer-vision skill acts as a comprehensive engineering companion for developing, training, and deploying visual AI systems. It provides deep technical guidance on object detection, semantic and instance segmentation, image classification, and video analysis. Designed for professional computer vision engineers, this skill leverages frameworks like PyTorch, Ultralytics, and Detectron2 to streamline the lifecycle of a vision model, from raw data preparation and architecture selection to production-grade deployment via TensorRT or ONNX. It effectively handles complex tasks such as model quantization, pruning, and infrastructure setup for real-time inference on edge or cloud platforms.

Installation

To integrate this capability into your agent workflow, use the package manager command: clawhub install openclaw/skills/skills/alirezarezvani/senior-computer-vision Ensure your local environment has PyTorch, OpenCV, and relevant dependencies installed before triggering the initial configuration scripts provided in the skill documentation.

Use Cases

• Real-time Surveillance: Implementing YOLOv8/v11 for high-speed object detection in live video streams.
• Medical Imaging: Configuring Mask R-CNN or SAM for accurate instance segmentation of anomalies in radiological data.
• Autonomous Systems: Designing lightweight inference pipelines optimized with TensorRT for deployment on embedded edge devices like NVIDIA Jetson.
• Automated Quality Control: Building classification and detection pipelines to identify manufacturing defects on assembly lines.
• Performance Tuning: Using the skill to profile model bottlenecks and apply quantization to balance precision and speed.

Example Prompts

1. "I need to deploy a YOLOv8 model for real-time detection on an NVIDIA Jetson Orin. Walk me through the steps for exporting to TensorRT and optimizing the precision for latency reduction."
2. "Compare the trade-offs between Faster R-CNN and RT-DETR for a high-accuracy, low-latency industrial inspection task. Which architecture would perform better with a dataset of 5,000 images?"
3. "Help me restructure my MMDetection config file to include custom albumentations for synthetic noise augmentation to improve model robustness against outdoor lighting variability."

Tips & Limitations

• Hardware Requirements: Many of the optimization tasks (TensorRT/quantization) require NVIDIA GPU access to generate calibrated binaries.
• Data Quality: This skill assumes you have or are building a high-quality annotated dataset. Always use tools like CVAT or Label Studio for consistency.
• Versioning: Ensure your environment matches the specific framework versions (e.g., PyTorch 2.x) to prevent dependency conflicts during compilation or deployment.