PayamFard123/dermaintel-face-parsing

BiSeNet Face Parsing Model (Hair Segmentation)

This is a BiSeNet-based face parsing model trained on CelebAMask-HQ dataset. Used for accurate hair segmentation in the DermaIntel application.

Model Details

• Architecture: BiSeNet with ResNet-18 backbone
• Format: ONNX
• Input Size: 512x512 RGB
• Classes: 19 facial components

Class Labels

Index	Label
0	background
1	skin
2	l_brow
3	r_brow
4	l_eye
5	r_eye
6	eye_g (glasses)
7	l_ear
8	r_ear
9	ear_r (earring)
10	nose
11	mouth
12	u_lip
13	l_lip
14	neck
15	neck_l
16	cloth
17	hair
18	hat

Usage

from huggingface_hub import hf_hub_download
import onnxruntime as ort

# Download model
model_path = hf_hub_download(
    repo_id="PayamFard123/dermaintel-face-parsing",
    filename="resnet18.onnx"
)

# Load model
session = ort.InferenceSession(model_path)

Original Source

Model weights from: https://github.com/yakhyo/face-parsing