app.py

from flask import Flask, request, jsonify
from flask_cors import CORS
import threading
import time
import traceback
import cv2, mediapipe as mp, numpy as np, base64, io
from PIL import Image

app = Flask(__name__)
CORS(app)

pose = None
mp_drawing = None
mp_pose = mp.solutions.pose

print("🚀 Flask app starting...")

@app.route("/", methods=["GET"])
def home():
    return "<h1>✅ Virtual Try-On API is Running (Health OK)</h1>", 200

@app.route("/health", methods=["GET"])
def health():
    return jsonify({"status": "ok"}), 200


def init_mediapipe():
    """Load MediaPipe Pose in a background thread."""
    global pose, mp_drawing
    try:
        print("🧠 Initializing MediaPipe Pose (CPU)...")
        start = time.time()
        pose = mp_pose.Pose(static_image_mode=False, min_detection_confidence=0.5)
        mp_drawing = mp.solutions.drawing_utils
        print("✅ MediaPipe initialized in", round(time.time() - start, 2), "seconds")
    except Exception as e:
        print("❌ MediaPipe init error:", e)
        traceback.print_exc()


# Start loading MediaPipe asynchronously
threading.Thread(target=init_mediapipe, daemon=True).start()


def overlay_dress(frame, dress, landmarks):
    if landmarks is None or dress is None:
        return frame

    h, w, _ = frame.shape

    def to_pixel(lm):
        return int(lm.x * w), int(lm.y * h)

    try:
        left_shoulder = to_pixel(landmarks[mp_pose.PoseLandmark.LEFT_SHOULDER.value])
        right_shoulder = to_pixel(landmarks[mp_pose.PoseLandmark.RIGHT_SHOULDER.value])
        left_hip = to_pixel(landmarks[mp_pose.PoseLandmark.LEFT_HIP.value])
        right_hip = to_pixel(landmarks[mp_pose.PoseLandmark.RIGHT_HIP.value])

        # calculate region for dress
        dress_width = int(np.linalg.norm(np.array(left_shoulder) - np.array(right_shoulder)) * 1.8)
        top_y = min(left_shoulder[1], right_shoulder[1])
        bottom_y = max(left_hip[1], right_hip[1])
        dress_height = int((bottom_y - top_y) * 1.2)

        # clamp values within frame bounds
        center_x = (left_shoulder[0] + right_shoulder[0]) // 2
        x1 = max(center_x - dress_width // 2, 0)
        y1 = max(top_y - 30, 0)
        x2 = min(x1 + dress_width, w)
        y2 = min(y1 + dress_height, h)

        # ✅ guard against invalid or tiny dimensions
        if x2 <= x1 or y2 <= y1:
            print("⚠️ Skipping overlay — invalid bounding box.")
            return frame

        # ✅ guard against zero-size dress
        if dress_width <= 0 or dress_height <= 0:
            print("⚠️ Skipping overlay — invalid dress size.")
            return frame

        dress_resized = cv2.resize(dress, (x2 - x1, y2 - y1), interpolation=cv2.INTER_AREA)

        # alpha blend if transparent
        if dress_resized.shape[2] == 4:
            alpha_s = dress_resized[:, :, 3] / 255.0
            alpha_l = 1.0 - alpha_s
            for c in range(3):
                frame[y1:y2, x1:x2, c] = (
                    alpha_s * dress_resized[:, :, c] +
                    alpha_l * frame[y1:y2, x1:x2, c]
                )
        else:
            # no transparency → just paste
            frame[y1:y2, x1:x2] = cv2.addWeighted(
                frame[y1:y2, x1:x2], 0.5, dress_resized, 0.5, 0
            )

    except Exception as e:
        print("⚠️ overlay_dress error:", e)
        traceback.print_exc()

    return frame



@app.route("/tryon", methods=["POST"])
def tryon():
    try:
        global pose
        if pose is None:
            return jsonify({"error": "Model still loading, please retry in 1–2 minutes"}), 503

        data = request.json.get("image")
        dress_data = request.json.get("dress")

        if not data or not dress_data:
            return jsonify({"error": "Missing image or dress data"}), 400

        img_bytes = base64.b64decode(data.split(",")[1])
        img = Image.open(io.BytesIO(img_bytes)).convert("RGB")
        frame = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)

        dress_bytes = base64.b64decode(dress_data.split(",")[1])
        dress_img = cv2.imdecode(np.frombuffer(dress_bytes, np.uint8), cv2.IMREAD_UNCHANGED)

        results = pose.process(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
        landmarks = results.pose_landmarks.landmark if results.pose_landmarks else None

        frame = overlay_dress(frame, dress_img, landmarks)
        if results.pose_landmarks:
            mp_drawing.draw_landmarks(frame, results.pose_landmarks, mp_pose.POSE_CONNECTIONS)

        _, buffer = cv2.imencode(".jpg", frame)
        img_base64 = "data:image/jpeg;base64," + base64.b64encode(buffer).decode()

        return jsonify({"image": img_base64})
    except Exception as e:
        print("❌ /tryon error:", e)
        traceback.print_exc()
        return jsonify({"error": str(e)}), 500


if __name__ == "__main__":
    import os
    port = int(os.environ.get("PORT", 7860))  # 👈 use 7860 (Hugging Face default)
    app.run(host="0.0.0.0", port=port)