app.py

import streamlit as st
import torch
import torch.nn as nn
from torchvision import transforms
from PIL import Image
import numpy as np
from tensorflow.keras.models import load_model
from torchvision.models import resnet18
import os
import requests
import mediapipe as mp
import cv2

# App title
st.title("🧠 Stroke Patient Pain Intensity Detector")

# Instructions
st.markdown(
    """
    Upload a full-face image of a stroke patient.
    The app will detect the **affected facial side** using a stroke classification model,
    and then use the **unaffected side** to predict **pain intensity** (PSPI score).
    """
)
st.write("🔧 Initializing and downloading models...")

# Download and load models
@st.cache_resource
def download_models():
    model_urls = {
        "cnn_stroke_model.keras": "https://huggingface.co/AdhamQQ/cnn_stroke_model/resolve/main/cnn_stroke_model.keras",
        "pain_model.pth": "https://huggingface.co/AdhamQQ/cnn_stroke_model/resolve/main/pain_model.pth"
    }
    for filename, url in model_urls.items():
        if not os.path.exists(filename):
            st.write(f"📥 Downloading {filename}...")
            r = requests.get(url)
            with open(filename, "wb") as f:
                f.write(r.content)
            st.success(f"✅ {filename} downloaded.")
        else:
            st.write(f"✔️ {filename} already exists.")

    stroke_model = load_model("cnn_stroke_model.keras")
    pain_model = resnet18(weights=None)
    pain_model.fc = nn.Linear(pain_model.fc.in_features, 1)
    pain_model.load_state_dict(torch.load("pain_model.pth", map_location=torch.device("cpu")))
    pain_model.eval()

    return stroke_model, pain_model

stroke_model, pain_model = download_models()

# Preprocessing for pain model
transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])

# MediaPipe Face Detection
mp_face = mp.solutions.face_detection
mp_draw = mp.solutions.drawing_utils

# Upload UI
uploaded_file = st.file_uploader("📂 Upload a full-face image", type=["jpg", "jpeg", "png"])

if uploaded_file is not None:
    st.write("📷 Image uploaded. Detecting face...")
    full_image = Image.open(uploaded_file).convert("RGB")
    img_np = np.array(full_image)

    with mp_face.FaceDetection(model_selection=1, min_detection_confidence=0.6) as detector:
        results = detector.process(cv2.cvtColor(img_np, cv2.COLOR_RGB2BGR))

        if not results.detections:
            st.error("❌ No face detected. Please upload a clear frontal face image.")
            st.stop()

        # Use first detection
        detection = results.detections[0]
        bboxC = detection.location_data.relative_bounding_box
        ih, iw, _ = img_np.shape
        x = int(bboxC.xmin * iw)
        y = int(bboxC.ymin * ih)
        w = int(bboxC.width * iw)
        h = int(bboxC.height * ih)

        face_crop = full_image.crop((x, y, x + w, y + h))
        st.image(full_image, caption="Uploaded Full-Face Image", use_column_width=True)

        # Split halves (face POV)
        fw, fh = face_crop.size
        fmid = fw // 2
        patient_right = face_crop.crop((0, 0, fmid, fh))    # viewer's left
        patient_left = face_crop.crop((fmid, 0, fw, fh))    # viewer's right

        # Stroke prediction input
        _, H, W, C = stroke_model.input_shape
        stroke_input = face_crop.resize((W, H))
        stroke_array = np.array(stroke_input).astype("float32") / 255.0
        stroke_array = np.expand_dims(stroke_array, axis=0)

        st.write("🧠 Predicting affected side of the face...")
        stroke_pred = stroke_model.predict(stroke_array)
        stroke_raw = stroke_pred[0][0]
        affected = int(np.round(stroke_raw))  # 0 = left affected, 1 = right affected

        if affected == 0:
            affected_side = "left"
            unaffected_side = "right"
            unaffected_face = patient_right
        else:
            affected_side = "right"
            unaffected_side = "left"
            unaffected_face = patient_left

        # Pain prediction
        st.write("📈 Predicting PSPI pain score from unaffected side...")
        input_tensor = transform(unaffected_face).unsqueeze(0)
        with torch.no_grad():
            output = pain_model(input_tensor)
            raw_score = output.item()
            pspi_score = max(0.0, min(raw_score, 6.0))

        # Display results
        st.subheader("🔍 Prediction Results")
        st.image(unaffected_face, caption="Unaffected Side Used for Pain Detection", width=300)
        st.write(f"**🧭 Affected Side (face POV):** `{affected_side}`")
        st.write(f"**✅ Unaffected Side (face POV):** `{unaffected_side}`")
        st.write(f"**🎯 Predicted PSPI Pain Score:** `{pspi_score:.3f}`")
        st.write(f"**📈 Raw Pain Model Output:** `{raw_score:.3f}`")
        st.write(f"**📊 Stroke Model Raw Output:** `{stroke_raw:.4f}`")

        st.markdown(
            """
            ---
            ### ℹ️ Stroke Model Output
            - Output is between `0` and `1`
            - Closer to `0` = Left side is affected
            - Closer to `1` = Right side is affected
            
            ### ℹ️ PSPI Score Scale
            - `0`: No pain  
            - `1–2`: Mild pain  
            - `3–4`: Moderate pain  
            - `5–6`: Severe pain
            """
        )