Sam-Vision/medical-skin-anatomy-segmentatio

Medical Skin Anatomy Segmentation - Training Version 5

Developed by: Sam Vision

💖 Support my work on Patreon

If you find this model useful, please consider supporting me on Patreon to help cover cloud GPU computing costs for training future versions (like the upcoming V6)!

A YOLOv8m-seg (instance segmentation) model trained to identify 15 anatomical body regions and coverage states.

Note: V5 in the filename (yolo_v5_best.pt) refers to the 5th training iteration of this project, NOT the YOLOv5 architecture. The actual architecture is YOLOv8m-seg from Ultralytics.

Architecture

Property	Value
Architecture	YOLOv8m-seg (Medium)
Task	Instance Segmentation
Developer	Sam Vision
Training Version	V5 (5th iteration)
Input Size	640x640
Confidence Threshold	0.35 (recommended)
Number of Classes	15
Framework	PyTorch / Ultralytics

Warning on Class Names

The class names listed in this model card are NOT the original training labels. They have been renamed to neutral/medical terminology for platform compliance purposes. The actual model internally uses different class names related to explicit human anatomy detection. If you load the model weights directly (.pt or .onnx), the original class names will be visible via model.names.

Classes (15 total)

Index	Published Name
0	body_f
1	pelvis_m
2	face_f
3	pelvis_f
4	gluteus_f
5	body_m
6	chest_f
7	lower_pelvis_f
8	covered_chest_f
9	covered_pelvis_f
10	face_m
11	covered_gluteus_f
12	chest_m
13	gluteus_m
14	lower_pelvis_m

To retrieve the actual class names, load the model directly:

from ultralytics import YOLO
model = YOLO('yolo_v5_best.pt')
print(model.names)  # Returns the original class names used during training

Usage

from ultralytics import YOLO
from huggingface_hub import hf_hub_download

model_path = hf_hub_download(
    repo_id='yujigiovani/medical-skin-anatomy-segmentatio',
    filename='yolo_v5_best.pt'
)
model = YOLO(model_path)
results = model('your_image.jpg', conf=0.35)
results[0].show()

Files

• yolo_v5_best.pt - PyTorch weights (recommended for training/fine-tuning)
• yolo_v5_best.onnx - ONNX export (recommended for inference/deployment, e.g. AnyLabeling)