import gradio as gr
from huggingface_hub import from_pretrained_keras
from PIL import Image
import numpy as np

# Load models
idpred = from_pretrained_keras("aegishield/idpred")
fingpred = from_pretrained_keras("aegishield/fingpred")

def show_fingername(fingernum):
    if fingernum>=5:
        fingername = "right "
        fingernum -= 5
    else: fingername = "left "
    if fingernum==0:
        fingername += "little"
    elif fingernum==1:
        fingername += "ring"
    elif fingernum==2:
        fingername += "middle"
    elif fingernum==3:
        fingername += "index"
    else: fingername += "thumb"
    return fingername

def predict_image(img):
    # Ensure the image is a PIL Image
    if not isinstance(img, Image.Image):
        img = Image.fromarray(img)

    # Resize the image and convert to grayscale
    img_resized = img.resize((96, 96)).convert('L')

    # Convert the resized grayscale image to a numpy array
    img_array = np.array(img_resized)

    # Add a channel dimension since the model expects (96, 96, 1)
    img_array = np.expand_dims(img_array, axis=-1)

    # Add batch dimension
    img_array = np.expand_dims(img_array, axis=0)

    # Predictions
    y_SubjectID_pred = idpred.predict(img_array)
    y_fingerNum_pred = fingpred.predict(img_array)

    # Extract prediction and confidence
    subject_id = np.argmax(y_SubjectID_pred, axis=1)[0] + 1
    finger_num = np.argmax(y_fingerNum_pred, axis=1)[0]
    subject_confidence = np.max(y_SubjectID_pred) * 100
    finger_confidence = np.max(y_fingerNum_pred) * 100

    output_message = f"Predicted Subject ID: {subject_id} (Confidence: {subject_confidence:.2f}%)\nPredicted Finger Type: {finger_num} (Confidence: {finger_confidence:.2f}%)"

    return output_message

# Create Gradio interface
iface = gr.Interface(fn=predict_image, inputs="image", outputs=["text", "text"])

# Launch interface
iface.launch()