YOLO v12 v26 Segmentation Edge — Benchmark Results

1 device(s) • 48 trained • 104 inference runs

Training Summary

Device	Exp	Arch	Size	Task	Approach	Batch	Epochs	Time	Best Ep.	mAP50(B)	mAP50(M)	mAP50-95(B)	Precision	Recall	F1	Conf

mAP50-95 by Model

Training Time vs Accuracy

Best F1 Score by Model

Optimal Confidence Threshold

mAP50 (Box) — Training Curves

mAP50-95 (Box) — Training Curves

mAP50 (Mask) — Training Curves

mAP50-95 (Mask) — Training Curves

Training Loss (Box + Seg + Cls)

Validation Loss (Box + Seg + Cls)

Precision (Box)

Recall (Box)

Convergence Analysis

Best epoch, early stopping, and efficiency metrics

Device	Model	Arch	Size	Approach	Total Ep.	Best Ep.	Patience Used	Best mAP50-95	Time	sec/epoch

Best Epoch vs Model Size

Training Efficiency (mAP50-95 / hour)

Inference Results

Device	Exp	Arch	Size	Approach	Format	Prec	ImgSz	Batch	Inf (ms)	Total (ms)	FPS	Median	p95	mAP50	mAP50-95	Model MB	GPU MB

FPS by Model

Latency Breakdown (ms)

YOLOv26 vs YOLOv12 — mAP50-95 by Size

YOLOv26 vs YOLOv12 — Best Epoch

Scratch vs Pretrained — mAP50-95

AutoBatch Size by Model

Per-Class Accuracy

Per-class mAP50 and mAP50-95 for each trained model. Sorted by class frequency (most to least common).

Model	Device

mAP50 by Class

mAP50-95 by Class

Hardware & Model Resources

Model complexity, GPU memory usage during training, and file sizes

Model	Device	Layers	Params	GFLOPs	GPU	Train GPU Mem	AutoBatch Mem	AutoBatch %	Weights (MB)