new_app.py

import streamlit as st
import keras
import numpy as np
from PIL import Image
import io, urllib.request, os

st.set_page_config(layout="wide")


@st.cache_data(show_spinner=False, ttl=600)
def fetch_satellite_tile(lat, lng, zoom=16, size="640x640", api_key=""):
    if not api_key:
        raise RuntimeError("Missing Google Static Maps API key in env var 'goog_api'.")
    url = (
       f"https://maps.googleapis.com/maps/api/staticmap?"
       f"center={lat},{lng}&zoom={zoom}&size={size}&maptype=satellite&key={api_key}"
    )
    buffer = io.BytesIO(urllib.request.urlopen(url, timeout=10).read())
    return Image.open(buffer).convert("RGB")


@st.cache_resource
def get_model():
    # compile=False skips optimizer/state rebuild, which saves time
    return keras.models.load_model("0.0008-0.92.keras", compile=False)

st.markdown("""
<style>
.block-container {
    padding-top: 2.2rem;
    padding-bottom: 0rem;
    padding-left: 5rem;
    padding-right: 5rem;
}
hr { 
    margin-top: 0.0cm;
    margin-bottom: 0.25cm;
}
[data-testid="stImage"] {
    display: flex !important;
    justify-content: center !important;
}
[data-testid="stImage"] img {
    margin: auto;
    display: block;
}
</style>
""", unsafe_allow_html=True)


#title
col1, col2 = st.columns(2)
with col1:
    _, col = st.columns(2)
    with col:
        st.markdown(
            "<h1 style='text-align: center; margin-bottom: 0.1em;'>Overpass Detector</h1>",
            unsafe_allow_html=True
        )
        
        st.markdown(
            """
            <div style="
                text-align:center;
                font-size:1.25rem;
                margin-top:-0.3rem;
                margin-bottom:0.6rem;
                opacity:0.9;
            ">
            This page runs a trained machine learning model to detect road overpasses in satellite imagery.
            </div>
            """,
            unsafe_allow_html=True
        )

with col2:
    _, col, _ = st.columns(3)
    with col:
        st.image('overpass.png')

st.write("---")

#load model and initialize image size required by model. uploaded images are resized to indicated size
img_height = 640
img_width = 640

state = st.session_state
state.loaded_model = get_model()

if "lat" not in state:
    state.lat = 39.11

if "lng" not in state:
    state.lng = -86.56 

if "coords_submitted" not in state:
    state.coords_submitted = False

if "img" not in state:
    state.img = None

# Preload default image once
if state.img is None:
    try:
        api_key = os.getenv("goog_api", "")
        state.img = fetch_satellite_tile(state.lat, state.lng, api_key=api_key)
    except Exception as e:
        st.info(f"Couldn’t fetch default tile: {e}")

col1, col2, col3 = st.columns([0.8, 1.8, 1.0])  # adjust ratios to taste

with col3:
    st.subheader('Enter latitude/longitude coordinates:')
    with st.form("coords_form"):
        c1, c2 = st.columns(2)
        with c1:
            st.number_input('Latitude', key="lat", min_value=-90.0, max_value=90.0, step=0.01, format="%.2f")
            st.write('The current lat/long are:')
        with c2:
            st.number_input('Longitude', key="lng", min_value=-180.0, max_value=180.0, step=0.01, format="%.2f")
            st.write(f"{st.session_state.lat:.2f}, {st.session_state.lng:.2f}")
        submitted = st.form_submit_button("Get Image and Prediction")

if submitted:
    try:
        api_key = os.getenv("goog_api", "")
        state.img = fetch_satellite_tile(st.session_state.lat, st.session_state.lng, api_key=api_key)
        st.rerun()  # ensures the updated lat/lng render immediately
    except Exception as e:
        st.error(f"Error fetching image: {e}")
    
with col2:
    if state.coords_submitted:
        state.coords_submitted = False
        try:
            api_key = os.getenv("goog_api", "")
            state.img = fetch_satellite_tile(state.lat, state.lng, api_key=api_key)
        except Exception as e:
            st.error(f"Error fetching image: {e}")

    if state.img is not None:
        st.markdown(
            "<div style='display:flex; justify-content:center;'>",
            unsafe_allow_html=True
        )
        st.image(state.img, width=640)
        st.markdown("</div>", unsafe_allow_html=True)

with col1:
    if state.img is not None:
        img_array = np.array(state.img)
        batch_size = 1
        img_array = np.reshape(img_array,[batch_size,img_height,img_width,3])
        with st.spinner("Running inference..."):
            result = state.loaded_model.predict(img_array)
        crossing_chance = result[0][1]*100
        pct = np.round(crossing_chance,decimals=2)
        
        label = "Overpass Likely" if pct >= 50 else "Overpass Unlikely"
        
        st.markdown(
            f"""
            <div style="text-align:center;">
                <div style="font-size:2.2rem; font-weight:700; margin-bottom:0.4rem;">
                    {label}
                </div>
                <div style="font-size:1.35rem;">
                    Model probability: <b>{pct}%</b>
                </div>
            </div>
            """,
            unsafe_allow_html=True
)