LanluZ/vascular-bundle-yolov10

Vascular Bundle Detection (YOLOv10m)

A YOLOv10m object-detection model for vascular bundles / bamboo cells in microscopy video frames. Trained on the data.yaml dataset from the companion GitHub repo LanluZ/vascular-bundle-track (formerly yolov8_test).

Model

• Architecture: YOLOv10m via Ultralytics 8.3.87
• Input size: 640
• Single class: Vascular

Metrics

Independent validation on data.yaml (2 images / 224 instances):

Precision	Recall	mAP50	mAP50-95	F1
0.9722	0.9777	0.9931	0.9765	0.9749

Getting started

from ultralytics import YOLO

model = YOLO("https://huggingface.co/LanluZ/vascular-bundle-yolov10/resolve/main/weights/best.pt")
results = model.predict("videos/56-fire.mp4", conf=0.58)

Or download the weight to a local folder and point Ultralytics at it:

hf download LanluZ/vascular-bundle-yolov10 weights/best.pt --local-dir .

Training config

• model: yolov10m.pt, epochs: 50, batch: 8, imgsz: 640, patience: 30
• optimizer: auto, seed: 0, deterministic: true
• Data: data.yaml (1 class: Vascular)

Files

File	Description
weights/best.pt	Final model (mAP50 0.9931)
weights/best_previous.pt	Previous best (mAP50 0.9830)

Related

• Source repo: LanluZ/vascular-bundle-track
• Source/demo videos: LanluZ/vascular-bundle-media