Computer Vision

This skill kicks in when you're building image classification, object detection, or segmentation pipelines. It emphasizes starting with pretrained models (ImageNet, COCO weights) and fine-tuning instead of training from scratch, using aggressive augmentation via albumentations, and separating training from inference optimization. You get concrete PyTorch examples for fine-tuning EfficientNet classifiers and running YOLO detection, plus guidance on when to use anchor-free architectures, focal loss for detection heads, and Dice loss for segmentation. The core philosophy is sound: validate on deployment-like data, monitor for distribution shift in production, and reach for foundation models like SAM or DINOv2 before committing to full training runs.

Install to Claude Code

npx -y skills add absolutelyskilled/absolutelyskilled --skill computer-vision --agent claude-code