app5.py

import streamlit as st
from PIL import Image
from ultralytics import YOLO
import cv2, os, logging
from datetime import datetime
import numpy as np

# ---------------- Config générale ----------------
MODEL_PATH = "best.pt"
SAVE_DIR = os.path.join("/tmp", "results")
os.makedirs(SAVE_DIR, exist_ok=True)

# ---------------- Limites version gratuite ----------------
SAVE_LIMIT_FREE = 5  # nombre max d'uploads pour version gratuite
uploads_count = 0
premium_access = False  # actif après clé premium

# ---------------- Logger ----------------
logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s")
logger = logging.getLogger("stroke-api")

# ---------------- Charger modèle YOLO ----------------
model = YOLO(MODEL_PATH)

# ---------------- Fonctions utilitaires ----------------
def _largest_face_bbox(np_img, min_conf=0.6):
    import mediapipe as mp
    mp_face_detection = mp.solutions.face_detection
    h, w = np_img.shape[:2]
    with mp_face_detection.FaceDetection(min_detection_confidence=min_conf) as fd:
        results = fd.process(cv2.cvtColor(np_img, cv2.COLOR_RGB2BGR))
        if not results.detections:
            return None
        boxes = []
        for det in results.detections:
            rel = det.location_data.relative_bounding_box
            x1 = int(max(0, rel.xmin) * w)
            y1 = int(max(0, rel.ymin) * h)
            x2 = int(min(1.0, rel.xmin + rel.width) * w)
            y2 = int(min(1.0, rel.ymin + rel.height) * h)
            boxes.append((x1, y1, x2, y2))
        boxes.sort(key=lambda b: (b[2]-b[0])*(b[3]-b[1]), reverse=True)
        return boxes[0] if boxes else None

def predict_image(image, conf=0.85, show_labels=True):
    global uploads_count, premium_access
    # Limite version gratuite
    if not premium_access:
        if uploads_count >= SAVE_LIMIT_FREE:
            st.warning(f"⚠️ Limite gratuite atteinte ({SAVE_LIMIT_FREE} images). Entrez une clé API pour continuer.")
            return None
        uploads_count += 1

    np_img = np.array(image)
    if np_img.shape[2] == 4:
        np_img = cv2.cvtColor(np_img, cv2.COLOR_RGBA2BGR)
    else:
        np_img = cv2.cvtColor(np_img, cv2.COLOR_RGB2BGR)

    # 1) Détection visage
    face_bbox = _largest_face_bbox(np_img)
    if face_bbox is None:
        st.warning("⚠️ Aucun visage humain détecté. Veuillez centrer le visage.")
        return None

    # 2) YOLO inference
    results = model.predict(source=np_img, conf=conf, verbose=False)

    # Vérifier si on a une détection valide
    if len(results[0].boxes) == 0:
        return None  # aucune détection au seuil demandé

    annotated_image = results[0].plot(labels=show_labels)
    out_path = os.path.join(SAVE_DIR, f"image_result_{datetime.now().strftime('%Y%m%d_%H%M%S')}.png")
    cv2.imwrite(out_path, annotated_image)
    return out_path

def predict_video(video_path, conf=0.85, show_labels=True):
    global uploads_count, premium_access
    if not premium_access:
        if uploads_count >= SAVE_LIMIT_FREE:
            st.warning(f"⚠️ Limite gratuite atteinte ({SAVE_LIMIT_FREE} vidéos). Entrez une clé API pour continuer.")
            return None
        uploads_count += 1

    cap = cv2.VideoCapture(video_path)
    fourcc = cv2.VideoWriter_fourcc(*'mp4v')
    out_path = os.path.join(SAVE_DIR, f"video_result_{datetime.now().strftime('%Y%m%d_%H%M%S')}.mp4")
    fps = cap.get(cv2.CAP_PROP_FPS) or 30
    width, height = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)), int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
    out = cv2.VideoWriter(out_path, fourcc, fps, (width, height))
    detections = 0

    while cap.isOpened():
        ret, frame = cap.read()
        if not ret:
            break
        # 1) Détection visage
        face_bbox = _largest_face_bbox(frame)
        if face_bbox is None:
            continue  # skip frame sans visage

        results = model.predict(frame, conf=conf, verbose=False)
        if len(results[0].boxes) > 0:
            detections += 1
        annotated = results[0].plot(labels=show_labels)
        out.write(annotated)

    cap.release()
    out.release()
    if detections == 0:
        return None
    return out_path

# ---------------- Interface Streamlit ----------------
st.title("🧠 Stroke-IA – Détection AVC par IA")

# Sidebar
st.sidebar.header("⚙️ Paramètres")
conf_threshold = st.sidebar.slider("Seuil de confiance", 0.1, 1.0, 0.85, 0.05)
show_labels = st.sidebar.checkbox("Afficher les labels", value=True)

st.sidebar.header("🔑 Mode Premium")
key_input = st.sidebar.text_input("Entrez votre clé premium")
if key_input == "VOTRE_CLE_PREMIUM":
    premium_access = True
    st.sidebar.success("✅ Mode premium activé !")
    key_input = ""  # effacer la saisie

# Section image upload
st.header("🖼️ Détection sur image")
image_file = st.file_uploader("Uploader une image", type=["jpg","jpeg","png"])
if image_file and st.button("Analyser l'image"):
    image = Image.open(image_file)
    result_path = predict_image(image, conf=conf_threshold, show_labels=show_labels)
    if result_path is None:
        st.success(f"✅ Aucune détection ou limite gratuite atteinte")
    else:
        st.image(result_path, caption="Image annotée", use_container_width=True)

# Section vidéo upload
st.header("🎥 Détection sur vidéo")
video_file = st.file_uploader("Uploader une vidéo (mp4, mov, etc.)", type=["mp4","mov"])
if video_file and st.button("Analyser la vidéo"):
    temp_path = os.path.join(SAVE_DIR, "temp_video.mp4")
    with open(temp_path,"wb") as f:
        f.write(video_file.read())
    result_path = predict_video(temp_path, conf=conf_threshold, show_labels=show_labels)
    if result_path is None:
        st.success(f"✅ Aucune détection ou limite gratuite atteinte")
    else:
        st.video(result_path)

# Disclaimer
st.markdown(f"""
---
👨‍💻 **Badsi Djilali** — Ingénieur Deep Learning  
🚀 Créateur de **Stroke_IA_Detection**  
🧠 (Détection d'asymétrie faciale & AVC par IA)

⚠️ **Disclaimer :** Stroke-IA est une démo technique, pas un avis médical.  
© {datetime.now().year} — Badsi Djilali.
""")