YOLO26-M-seg

Ultralytics YOLO26 is the latest evolution in the YOLO series of real-time object detectors, engineered from the ground up for edge and low-power devices. This is the segmentation variant that provides instance segmentation capabilities.

Model Specifications

Property	Value
Input Size	640 pixels
mAP Box (50-95, e2e)	52.5
mAP Mask (50-95, e2e)	44.1
CPU Speed (ONNX)	328.2 ms
T4 TensorRT10 Speed	6.7 ms
Parameters	23.6M
FLOPs	121.5B

Key Features

The architecture of YOLO26 is guided by three core principles:

Simplicity: YOLO26 is a native end-to-end model, producing predictions directly without the need for non-maximum suppression (NMS). By eliminating this post-processing step, inference becomes faster, lighter, and easier to deploy in real-world systems.

Deployment Efficiency: The end-to-end design cuts out an entire stage of the pipeline, dramatically simplifying integration, reducing latency, and making deployment more robust across diverse environments.

Training Innovation: YOLO26 introduces the MuSGD optimizer, a hybrid of SGD and Muon — inspired by Moonshot AI's Kimi K2 breakthroughs in LLM training. This optimizer brings enhanced stability and faster convergence.

Instance Segmentation Enhancements

• Semantic Segmentation Loss: Improves model convergence
• Multi-scale Proto Module: Leverages multi-scale information for superior mask quality
• DFL Removal: Simplified inference and broader hardware compatibility
• End-to-End NMS-Free Inference: Reduced latency and easier production integration

Usage

Install ultralytics with pip install ultralytics.

Download the model.

from huggingface_hub import hf_hub_download

model_path = hf_hub_download(repo_id="openvision/yolo26-m-seg", filename="model.pt")

Infer.

from ultralytics import YOLO
from PIL import Image
import requests

model = YOLO(model_path)

# Train the model on the COCO8 example dataset for 100 epochs
results = model.train(data="coco8.yaml", epochs=100, imgsz=640)

url = 'http://images.cocodataset.org/val2017/000000039769.jpg'
image = Image.open(requests.get(url, stream=True).raw)

# Run inference with the YOLO26m-seg model on the image
results = model.predict(image)

Documentation

For more information, see the official YOLO26 documentation.