app.py

import gradio as gr
import torchvision.transforms as transforms
from torchvision.transforms import InterpolationMode
import torch
from huggingface_hub import hf_hub_download

from model import Model

# Load Model
model_path = hf_hub_download(
    repo_id="itserr/exvoto_classifier_convnext_base_224",
    filename="model.pt"
)

model = Model('convnext_base')
ckpt = torch.load(model_path, map_location=torch.device("cpu"))  # Ensure compatibility
model.load_state_dict(ckpt['model'])

device = 'cuda' if torch.cuda.is_available() else 'cpu'
model.to(device)
model.eval()


# Image Transformations
transform = transforms.Compose([
    transforms.Resize(size=(224,224), interpolation=InterpolationMode.BICUBIC),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])

# Classification Function
def classify_img(img, threshold):
    classification_threshold = threshold
    img_tensor = transform(img).unsqueeze(0).to(device)

    with torch.no_grad():
        pred = model(img_tensor)
        score = torch.sigmoid(pred).item()
        
    # Determine Prediction
    if score >= classification_threshold:
        label = "✅ This is an **Ex-Voto** image!"
    else:
        label = "❌ This is **NOT** an Ex-Voto image."
    
    # Format Confidence Score
    confidence = f"The probability that the image is an ex-voto is: {score:.2%}"
    return label, confidence

example_images = [['examples/exvoto1.jpg', None],
                ['examples/exvoto2.jpg', None],
                ['examples/nonexvoto1.jpg', None],
                ['examples/nonexvoto2.jpg', None],
                ['examples/natural1.jpg', None],
                ['examples/natural2.jpg', None],]

# Function to Clear Outputs When a New Image is Uploaded
def clear_outputs(img):
    return gr.update(value=""), gr.update(value="")

# Gradio Interface
with gr.Blocks() as demo:
    gr.Markdown("## Ex-Voto Image Classifier")
    gr.Markdown("📸 **Upload an image** to check if it's an **Ex-Voto** painting!")
    
    with gr.Row():
        with gr.Column(scale=2):  # Left section: Image upload & slider
            img_input = gr.Image(type="pil")
            threshold_slider = gr.Slider(
                minimum=0.5, maximum=1.0, value=0.7, step=0.1, label="Classification Threshold"
            )
            submit_btn = gr.Button("Classify")

        with gr.Column(scale=1):  # Right section: Prediction & Confidence
            prediction_output = gr.Textbox(label="Prediction", interactive=False)
            confidence_output = gr.Textbox(label="Confidence Score", interactive=False)

    # Clear outputs when a new image is uploaded
    img_input.change(fn=clear_outputs, inputs=[img_input], outputs=[prediction_output, confidence_output])

    # Submit button triggers classification
    submit_btn.click(fn=classify_img, inputs=[img_input, threshold_slider], outputs=[prediction_output, confidence_output])

    # Example images (Only show images, no threshold value)
    gr.Examples(
        examples=example_images,
        inputs=[img_input]
    )

# Launch App
demo.launch()