main.py

import gradio as gr
import numpy as np
import cv2
import os
from huggingface_hub import hf_hub_download
import torch
import segmentation_models_pytorch as smp
import importlib.util
#import onnxruntime as ort

REPO_ID = "IFMedTech/Skin-Analysis"
# List of Python files and corresponding class names
PY_MODULES = {
    "dark_circles.py": "DarkCircleDetector",
    "inflammation.py": "RednessDetector",
    "texture.py": "TextureDetector",
    "skin_tone.py": "SkinToneDetector",
    "oiliness.py": "OilinessDetector", 
    "wrinkle_unet.py": "WrinkleDetector",
    "age.py": "AgePredictor"
}
# def load_model(token):
#     """Download and load ONNX model"""
#     model_path = hf_hub_download(
#         repo_id=REPO_ID,
#         filename="model/wrinkle_model.onnx.data",  # Adjust path if needed
#         token=token
#     )
    
#     # Create ONNX Runtime session
#     session = ort.InferenceSession(
#         model_path,
#         providers=['CUDAExecutionProvider', 'CPUExecutionProvider'] 
#         if torch.cuda.is_available() else ['CPUExecutionProvider']
#     )
    
#     device = "cuda" if torch.cuda.is_available() else "cpu"
#     return session, device
def load_model(token):
    repo_id = "IFMedTech/Skin-Analysis"
    filename = "model/wrinkles_unet_v1.pth"
    # token = os.environ.get("HUGGINGFACE_HUB_TOKEN")  # Set this env var with your token
    
    # if not token:
    #     raise ValueError("HUGGINGFACE_HUB_TOKEN environment variable is required for private repo access.")
    
    model_path = hf_hub_download(repo_id=repo_id, filename=filename, token=token)
    
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    model = smp.Unet(
        encoder_name="resnet34",
        encoder_weights=None,
        in_channels=3,
        classes=1
    )
    model.load_state_dict(torch.load(model_path, map_location=device))
    model.to(device)
    model.eval()
    return model, device

def dynamic_import(module_path, class_name):
    spec = importlib.util.spec_from_file_location(class_name, module_path)
    module = importlib.util.module_from_spec(spec)
    spec.loader.exec_module(module)
    return getattr(module, class_name)

# Dynamically download and import modules
detector_classes = {}
token = os.environ.get("HUGGINGFACE_TOKEN")
if not token:
    raise ValueError("Please set the HUGGINGFACE_TOKEN environment variable in repo secrets!")

for py_file, class_name in PY_MODULES.items():
    py_path = hf_hub_download(
        repo_id=REPO_ID,
        filename=py_file,
        token=token
    )
    detector_classes[class_name] = dynamic_import(py_path, class_name)

# --- Skin analysis function using downloaded detectors ---
def analyze_skin(image: np.ndarray, analysis_type: str) -> np.ndarray:
    output = image.copy()
    if analysis_type == "Dark Circles":
        detector = detector_classes["DarkCircleDetector"](image)
        result = detector.predict_json()
        output = detector.draw_json()
    elif analysis_type == "Redness":
        detector = detector_classes["RednessDetector"](image)
        result = detector.predict_json()
        output = result.get("overlay_image")
    elif analysis_type == "Texture":
        detector = detector_classes["TextureDetector"](image)
        result = detector.predict_json()
        # print(result)
        output = result.get("overlay_image")
    elif analysis_type == "Skin Tone":
        detector = detector_classes["SkinToneDetector"](image)
        result = detector.predict_json()
        output = result.get("output_image")
    elif analysis_type == "Oiliness":
        detector = detector_classes["OilinessDetector"](image)
        result = detector.predict_json()
        if result.get("detected"):
            output = result.get("overlay_image")
        #     print(f"Oiliness scores: {result.get('scores')}")
        # else:
        #     print(f"Oiliness detection error: {result.get('error')}")
    elif analysis_type == "Wrinkles":
        model, device = load_model(token)
        detector = detector_classes["WrinkleDetector"](image, model, device)
        result = detector.predict_json()
        if result.get("detected") is not None:
            output = detector.draw_json(result)

    elif analysis_type == "Skin Age":
        detector = detector_classes["AgePredictor"](image)
        result = detector.predict_json()
        output = detector.draw_json(result)
        
    return output


# --- Gradio Interface code ---
app = gr.Interface(
    fn=analyze_skin,
    inputs=[
        gr.Image(type="numpy", label="Upload your face image"),
        gr.Radio(
            ["Dark Circles", "Redness", "Texture", "Skin Tone", "Oiliness", "Wrinkles", "Skin Age"],  
            label="Select Skin Analysis Type"
        ),
    ],
    outputs=gr.Image(type="numpy", label="Analyzed Image"),
    title="Skin Analysis Demo",
    description="Upload an image and choose a skin analysis parameter."
)

if __name__ == "__main__":
    app.launch(server_name="0.0.0.0", server_port=7860)