Скачать или смотреть YOLOv26 for Small Object Detection & Segmentation in Custom Dataset with (

In this video, I demonstrate a complete step-by-step tutorial on training YOLOv26 for small object detection and segmentation using a custom agricultural dataset. The focus is on early-stage green apple fruitlets in commercial orchards, where objects are extremely small, dense, and visually complex.

The dataset was manually annotated by human experts in Roboflow and exported in YOLO format, with each fruitlet divided into three fine-grained classes: Calyx, Fruitlet, and Peduncle. Around 600 high-resolution orchard images were labeled, producing thousands of small-object instances, making this a realistic and challenging benchmark for small object detection.

This tutorial first trains YOLOv26-Segmentation (YOLO26n-seg) using standard full-image training, and then introduces a SAHI-ready training strategy with higher input resolution. SAHI (Slicing Aided Hyper Inference) improves detection by slicing large images into overlapping patches, allowing small objects to appear larger and more distinguishable during inference.

Key Results & Performance Gains

Without SAHI, YOLOv26 achieved moderate mAP for small classes. After adopting a SAHI-ready setup, we observed a significant improvement in precision, recall, and mAP, especially for the most challenging classes:

Overall mAP50 improved from ~0.49 to ~0.65

Fruitlet class mAP50 increased to ~0.77

Better segmentation accuracy for calyx and peduncle

Maintained real-time inference speed with a lightweight 6.6 MB model

The final model runs efficiently on GPU and is suitable for real-world deployment, including precision agriculture, yield estimation, and robotic perception.

📌 What you’ll learn in this video:

YOLOv26 environment setup

Custom dataset preparation (Roboflow → YOLO)

Small object-focused training strategy

SAHI-based inference for accuracy boost

Interpreting detection & segmentation metrics

If you’re working on YOLOv26, small object detection, segmentation, or SAHI-based inference, this tutorial will help you implement a robust and scalable pipeline.

👉 Don’t forget to subscribe for more advanced YOLOv26 and AI tutorials!