mindbox

Detect GPU hardware and CUDA environment, output a structured report for downstream skills. Use this skill before any GPU-intensive task (training, inference, rendering) to determine available hardware and choose appropriate model sizes, batch sizes, and device settings.

Fine-tune YOLO object detection models on custom datasets, export to ONNX, with TensorBoard logging. Use this skill whenever the user mentions object detection, YOLO fine-tuning, bounding box detection, 物体检测, 目标检测, 缺陷检测, or wants to detect/locate/identify objects in images — even if they don't explicitly say "YOLO" or "object detection". Also applies to industrial defect detection, medical imaging detection, autonomous driving perception, and security surveillance scenarios.