app.py

import streamlit as st
import cv2
import numpy as np
from ultralytics import YOLO
from PIL import Image

st.set_page_config(layout="wide")
st.title("OPG Segmentation + Midline + Sinus Detection")

# -----------------------------
# Load model (cached)
# -----------------------------
@st.cache_resource
def load_model():
    return YOLO("best.pt")

model = load_model()

# -----------------------------
# Upload image
# -----------------------------
uploaded_file = st.file_uploader("Upload OPG Image", type=["jpg", "png", "jpeg"])

# -----------------------------
# Preprocessing Function
# -----------------------------
def preprocess_image(image):

    # Convert to grayscale
    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

    # Apply CLAHE
    clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
    gray = clahe.apply(gray)

    # Resize so max dimension = 2048
    h, w = gray.shape
    scale = 2048 / max(h, w)
    gray = cv2.resize(gray, (int(w * scale), int(h * scale)))

    # Convert back to 3-channel (YOLO expects 3 channels)
    processed = cv2.cvtColor(gray, cv2.COLOR_GRAY2BGR)

    return processed



if uploaded_file is not None:

    # Convert to OpenCV format
    image = Image.open(uploaded_file).convert("RGB")
    image = np.array(image)
    image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)

    image = preprocess_image(image)

    h, w, _ = image.shape

    # -----------------------------
    # Run prediction
    # -----------------------------
    results = model(image, conf=0.25)
    result = results[0]

    tooth_centers_x = []

    # -----------------------------
    # Collect tooth centers
    # -----------------------------
    for box, cls in zip(result.boxes.xywh, result.boxes.cls):
        cls = int(cls)
        if cls == 0:  # tooth class
            x_center = box[0].item()
            tooth_centers_x.append(x_center)

    if len(tooth_centers_x) == 0:
        st.warning("No teeth detected!")
        st.stop()

    # -----------------------------
    # Compute midline
    # -----------------------------
    midline_x = int((min(tooth_centers_x) + max(tooth_centers_x)) / 2)

    # -----------------------------
    # Draw tooth masks
    # -----------------------------
    if result.masks is not None:
        for mask, cls in zip(result.masks.xy, result.boxes.cls):
            cls = int(cls)
            if cls == 0:
                polygon = np.array(mask, dtype=np.int32)
                cv2.polylines(image, [polygon], True, (0, 255, 0), 2)

    # -----------------------------
    # Process sinus
    # -----------------------------
    for box, cls in zip(result.boxes.xywh, result.boxes.cls):
        cls = int(cls)
        if cls == 1:  # sinus class
            x, y, bw, bh = box
            x_center = x.item()

            if x_center < midline_x:
                label = "Right Sinus"
                color = (255, 0, 0)
            else:
                label = "Left Sinus"
                color = (0, 0, 255)

            x1 = int(x - bw / 2)
            y1 = int(y - bh / 2)
            x2 = int(x + bw / 2)
            y2 = int(y + bh / 2)

            cv2.rectangle(image, (x1, y1), (x2, y2), color, 2)
            cv2.putText(image, label, (x1, y1 - 10),
                        cv2.FONT_HERSHEY_SIMPLEX, 0.7, color, 2)

    # -----------------------------
    # Draw midline
    # -----------------------------
    cv2.line(image, (midline_x, 0), (midline_x, h),
             (0, 255, 255), 2)

    # Convert back to RGB for display
    display_image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)

    col1, col2 = st.columns(2)

    with col1:
        st.image(uploaded_file, caption="Original Image")

    with col2:
        st.image(display_image, caption="Segmented Output")

    # -----------------------------
    # Download button
    # -----------------------------
    result_pil = Image.fromarray(display_image)
    st.download_button(
        label="Download Result Image",
        data=result_pil.tobytes(),
        file_name="output_with_midline.jpg",
        mime="image/jpeg"
    )