skills.rest
Tgoldi
TgoldiCOMMUNITY
Explore all skills by this author →
GitHub

senior-computer-vision

Builds object detection, segmentation, and deployment pipelines using PyTorch, YOLO, ONNX, and TensorRT.

1|Updated May 10, 2026
Search Tags:#computer vision#object detection#model optimization#onnx#yolo#image segmentation#tensorrt
One-click install
npx skills add https://github.com/Tgoldi/claude-skills --skill senior-computer-vision-tgoldi
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: senior-computer-vision
Source: https://github.com/Tgoldi/claude-skills/tree/main/senior-computer-vision
Command: npx skills add https://github.com/Tgoldi/claude-skills --skill senior-computer-vision-tgoldi

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires onnx, onnxruntime, torch, and includes scripts (resource) and references (resource) components.

What problem does it solve? Building production computer vision systems requires navigating architecture selection, dataset format conversion, training configuration, and inference optimization across many frameworks. This Skill provides structured workflows and automation scripts that take a vision project from raw annotated images to an optimized deployed model. ## Core Features & Use Cases - Object Detection Pipelines: Step-by-step guidance for training YOLOv8, Faster R-CNN, DETR, and RT-DETR models with Ultralytics, Detectron2, and MMDetection, including architecture selection tables based on speed and accuracy requirements. - Dataset Preparation: The dataset_pipeline_builder.py script analyzes datasets, detects annotation formats, converts between COCO, YOLO, and Pascal VOC, checks class imbalance, and generates augmentation configs and train/val/test splits. - Model Optimization and Deployment: The inference_optimizer.py script benchmarks PyTorch and ONNX models, reports latency percentiles and throughput, and recommends optimization paths such as TensorRT FP16/INT8, OpenVINO, and CoreML. - Use Case: You have 5,000 Pascal VOC-annotated images and need a real-time detector on an NVIDIA edge device. Use the dataset script to convert to COCO and split the data, train YOLOv8m following the detection workflow, then export to ONNX and TensorRT INT8 with the optimizer script. ## Quick Start Ask the assistant to analyze your image dataset and generate a YOLO training configuration using the dataset pipeline builder script.

Frequently Asked Questions about senior-computer-vision

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I train a YOLOv8 object detection model on a custom dataset?▼

Convert your annotations to COCO or YOLO format, create train/val/test splits, and generate a data.yaml config. Then run yolo detect train with your data file, a pretrained model like yolov8m.pt, and settings such as epochs=100 and imgsz=640.

How to convert Pascal VOC annotations to COCO format?▼

Use the dataset_pipeline_builder.py script with the convert command, pointing it at your VOC XML annotations and specifying COCO as the target format. It parses each XML file, extracts bounding boxes and class names, and writes a COCO annotations.json with categories.

YOLO vs Faster R-CNN for object detection, which should I choose?▼

YOLO is a single-stage detector suited for real-time needs above 30 FPS, while Faster R-CNN is a two-stage detector offering better localization at slower speeds around 46ms per image. Choose YOLO for edge or video applications and Faster R-CNN when accuracy matters more than latency.

How do I speed up PyTorch model inference for production deployment?▼

Export the model to ONNX, then convert to TensorRT with FP16 for NVIDIA GPUs or OpenVINO for Intel CPUs. The inference_optimizer.py script benchmarks baseline latency and recommends optimization paths, typically achieving 2-4x speedup with FP16 and 3-6x with INT8 quantization.

Does INT8 quantization hurt object detection accuracy?▼

INT8 quantization typically reduces mAP by 1-3% while shrinking model size by 75% and speeding inference 2-4x. It requires a calibration dataset of around 500 representative samples to minimize accuracy loss.

How many images do I need to fine-tune a detection model?▼

YOLO fine-tuning works with 100-500 images but 1,000-5,000 is recommended, while transformer detectors like DETR and DINO need 10,000 or more. For small datasets, use transfer learning with pretrained weights and aggressive augmentation.

Related Skills

Explore More Software Engineering Skills→
nelohenriqnelohenriq
1

segment-anything-model

Segment objects in images using point, box, or mask prompts.

Community
Intermediate
masermediagroup-stackmasermediagroup-stack
1

model-serving

Deploy LLM and ML models for production inference with vLLM, BentoML, and streaming APIs.

Community
Advanced
organvm-i-theoriaorganvm-i-theoria

ml-experiment-tracker

Track ML experiments, version datasets, and manage model registries with MLflow, W&B, and DVC.

Official
Intermediate
Orchestra-ResearchOrchestra-Research
11.5k

segment-anything-model

Segment any object in images using zero-shot foundation models.

Official
Intermediate