import gradio as gr
import numpy as np
import face_recognition

def compare_faces(image1, image2):
    img1 = np.array(image1)
    img2 = np.array(image2)
    img1_encodings = face_recognition.face_encodings(img1)
    img2_encodings = face_recognition.face_encodings(img2)
    
    if not img1_encodings or not img2_encodings:
        return "Face not detected in one or both images. Please try another image."
    
    distance = np.linalg.norm(img1_encodings[0] - img2_encodings[0])
    # Modify the similarity formula to make 60-70% into 90-100%
    similarity = 100 - (distance * 10)  # Amplify the impact of small distances

    # Apply a non-linear transformation to boost the 60-70% range to 90-100%
    if 60 <= similarity < 70:
        similarity = 90 + (similarity - 60) * (10 / 10)  # Linear map 60-70 to 90-100

    return f"Similarity: {similarity:.2f}%"

with gr.Blocks() as demo:
    gr.Markdown("### Face Similarity Checker")
    gr.Markdown("Upload two images to compare the faces in them.")
    gr.Markdown("- 70 - 80 % can be called as **identical**, even the same person with different position got this result")
    gr.Markdown("- 60 - 70 % can be called as **perfect couple**")
    gr.Markdown("This is documentation about results, I am not fixing this up to 9*% because model works well whether there is a face or not.")
    
    with gr.Row():
        with gr.Column():
            img1 = gr.Image(label="Upload Image 1", image_mode='RGB', type="pil")
        with gr.Column():
            img2 = gr.Image(label="Upload Image 2", image_mode='RGB', type="pil")
    btn = gr.Button("Compare Faces")
    output = gr.Label()
    
    btn.click(compare_faces, inputs=[img1, img2], outputs=output)

demo.launch()