Senior Computer Vision Engineer
This skill provides production-grade computer vision engineering guidance for object detection, image segmentation, and visual AI system deployment. It covers architecture selection, dataset preparation, model training, optimization, and deployment using frameworks like PyTorch, Ultralytics, Detectron2, and MMDetection.
✨ What it does
- Provides step-by-step workflows for detection, optimization, and dataset preparation
- Includes architecture selection guides with speed/accuracy trade-offs
- Offers scripts for training, inference optimization, and dataset pipeline building
- Covers export to ONNX, TensorRT, OpenVINO, CoreML, and quantization
- Recommends augmentations and split strategies for datasets
- Defines performance targets for real-time, high-accuracy, and edge scenarios
🎯 When to use it
- Building an object detection or segmentation pipeline from scratch
- Training custom vision models on custom datasets
- Optimizing and deploying trained models to production (ONNX, TensorRT, etc.)
- Preparing and cleaning custom datasets for vision training
- Selecting the right architecture for speed/accuracy/edge constraints
🚀 How to use
Trigger by asking for help with computer vision tasks like building a detection pipeline, training a YOLO model, optimizing inference, or preparing a dataset. Provide details such as target objects, dataset location, deployment target, and accuracy/speed requirements. Example prompts:
Help me build an object detection pipeline for detecting cars and pedestrians in real-time on an edge device.
I have a custom dataset in VOC format, how do I convert it to COCO and train a Mask R-CNN model?
📄 Output: The skill produces configuration files, scripts, and guidance for training, optimizing, and deploying computer vision models.
📦 Add this skill to Claude Code
# 1. Get the skills repo
git clone --depth 1 https://github.com/alirezarezvani/claude-skills /tmp/claude-skills
# 2. Copy this skill into your project (or ~/.claude/skills for all projects)
mkdir -p .claude/skills
cp -r /tmp/claude-skills/engineering-team/skills/senior-computer-vision .claude/skills/senior-computer-visionSkill source: engineering-team/skills/senior-computer-vision/SKILL.md
⚠️ Good to know
Requires familiarity with Python and the listed frameworks; assumes access to appropriate hardware for training and deployment.
❓ FAQ
What is the recommended architecture for real-time detection on edge devices?
YOLOv8n or MobileNetV3-SSD are recommended for edge deployment due to their lightweight nature and speed.
What is the typical accuracy drop when quantizing to INT8?
INT8 quantization typically results in a 1-3% accuracy drop compared to FP32.
How should I split my dataset for training?
For datasets under 1,000 images use 70/15/15; for 1,000-10,000 use 80/10/10; for over 10,000 use 90/5/5.
🤖 Overview, features, install steps and FAQ were generated from the project's SKILL.md on Sep 4, 2026. Always check the original source before running commands.