SYSTEM Cited by 1 source
YOLO (You Only Look Once)¶
YOLO (You Only Look Once) is a family of single-stage real-time object-detection models (YOLOv1 → YOLOv8 → YOLOv10+). Typical framing: a single neural-network pass produces bounding boxes + class labels + confidence scores over an entire image, in contrast to two-stage detectors (R-CNN family) that first propose regions then classify them. The single-stage design trades a small accuracy delta at the top end for order-of-magnitude inference speed, making YOLO the default choice for real-time + edge CV deployments.
Stub page — expand as more YOLO-internals sources land on the wiki.
Training notes (from ingested source)¶
- Cosine learning-rate scheduling + AdamW optimisation on PyTorch 2.1 named as "critical for stabilizing the larger YOLOv8l model variant and preventing gradient divergence during training" (sources/2026-04-01-aws-automate-safety-monitoring-with-computer-vision-and-generative-ai).
- YOLO annotation format (
.txtper image with normalised bounding-box coords + class index) is the canonical training input; GLIGEN-generated scenes are converted to this format by parallel Python workers downstream of synthetic-data generation. - Used for simultaneous multi-dimensional detection — in the AWS safety-monitoring PPE module, one YOLO model locates workers + classifies presence/absence of required safety equipment + applies contextual analysis to determine which PPE items are mandatory for the specific zone.
Seen in¶
- sources/2026-04-01-aws-automate-safety-monitoring-with-computer-vision-and-generative-ai — YOLOv8 + YOLOv8l models trained entirely on GLIGEN synthetic data reach 99.5% mAP@50 for PPE (3 classes) + 94.3% mAP@50 for Housekeeping (7 classes) without a single manually-annotated real image; PPE model hits 100% precision + 100% recall across all three classes. Served through SageMaker Serverful endpoints on ml.g6 instances.
Related¶
- patterns/synthetic-data-generation — the training-data production pattern paired with YOLO in the canonical instance.
- systems/gligen — the grounded generator producing YOLO- format-compatible training data.
- systems/aws-sagemaker-ai — the managed training + serving substrate.