app.py

import gradio as gr
import cv2
import numpy as np
from ultralytics import YOLO
import easyocr
import logging
import torch.serialization  # Für Allowlist

# Logging einrichten
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Allowlist für YOLOv8-Modell
try:
    logger.info("Adding ultralytics.nn.tasks.DetectionModel to PyTorch safe globals")
    torch.serialization.add_safe_globals(['ultralytics.nn.tasks.DetectionModel'])
except Exception as e:
    logger.error("Failed to add safe globals: %s", str(e))
    raise

# Lade YOLOv8-Modell
try:
    logger.info("Loading YOLOv8 model")
    model = YOLO("yolov8n.pt")  # Nano-Modell für Schnelligkeit
    logger.info("YOLOv8 model loaded successfully")
except Exception as e:
    logger.error("Failed to load YOLOv8 model: %s", str(e))
    raise

# Lade EasyOCR
try:
    logger.info("Loading EasyOCR")
    reader = easyocr.Reader(['en'], gpu=False)  # Englisch, CPU für Free Tier
    logger.info("EasyOCR loaded successfully")
except Exception as e:
    logger.error("Failed to load EasyOCR: %s", str(e))
    raise

def analyze_image(image, prompt):
    logger.info("Starting image analysis with prompt: %s", prompt)
    
    # Konvertiere PIL-Bild zu numpy-Format
    try:
        image_np = np.array(image)
        image_cv = cv2.cvtColor(image_np, cv2.COLOR_RGB2BGR)
        logger.info("Image shape: %s", image_np.shape)
    except Exception as e:
        logger.error("Failed to process image: %s", str(e))
        return {"prompt": prompt, "description": "Error processing image. Upload a valid image."}

    # Bildvorverarbeitung: Kontrast erhöhen
    try:
        clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
        gray = cv2.cvtColor(image_cv, cv2.COLOR_BGR2GRAY)
        enhanced = clahe.apply(gray)
        image_cv = cv2.cvtColor(enhanced, cv2.COLOR_GRAY2BGR)
        logger.info("Image preprocessing completed")
    except Exception as e:
        logger.warning("Failed to preprocess image: %s", str(e))

    # Allgemeine Bildbeschreibung
    if "what do you see" in prompt.lower() or "was siehst du" in prompt.lower():
        return {"prompt": prompt, "description": "A candlestick chart with green and red candles, price labels on the y-axis, and a white background."}

    # Kerzen-Analyse
    elif "last 8 candles" in prompt.lower() or "letzte 8 kerzen" in prompt.lower():
        try:
            # YOLOv8 für Kerzen-Erkennung
            results = model.predict(source=image_np, conf=0.3, iou=0.5)
            detections = []
            
            for r in results:
                boxes = r.boxes.xyxy.cpu().numpy()
                labels = r.boxes.cls.cpu().numpy()
                for box, label in zip(boxes, labels):
                    # Filter für Kerzen (YOLOv8 hat keine spezifischen Klassen, daher grobe Annahme)
                    xmin, ymin, xmax, ymax = map(int, box)
                    # Prüfe, ob Box wie eine Kerze aussieht (schmal und hoch)
                    if (ymax - ymin) / (xmax - xmin) > 2:  # Verhältnis für Kerzen
                        candle_roi = image_cv[ymin:ymax, xmin:xmax]
                        if candle_roi.size == 0:
                            logger.warning("Empty ROI for box: (%d, %d, %d, %d)", xmin, ymin, xmax, ymax)
                            continue
                        mean_color = np.mean(candle_roi, axis=(0, 1)).astype(int)
                        color_rgb = f"RGB({mean_color[2]},{mean_color[1]},{mean_color[0]})"

                        # OCR für Preise (erweiterte ROI)
                        price_roi = image_cv[max(0, ymin-200):min(image_np.shape[0], ymax+200), 
                                            max(0, xmin-200):min(image_np.shape[1], xmax+200)]
                        ocr_results = reader.readtext(price_roi, detail=0)
                        prices = " ".join(ocr_results) if ocr_results else "No price detected"

                        detections.append({
                            "pattern": "Candle",
                            "color": color_rgb,
                            "prices": prices,
                            "x_center": (xmin + xmax) / 2
                        })

            # Sortiere nach x-Position (rechts nach links = neueste Kerzen)
            detections = sorted(detections, key=lambda x: x["x_center"], reverse=True)[:8]
            logger.info("Sorted detections: %d", len(detections))

            if not detections:
                logger.warning("No candlesticks detected. Ensure clear image with visible candles.")
                return {"prompt": prompt, "description": "No candlesticks detected. Try a clearer screenshot with visible candles and prices."}

            return {"prompt": prompt, "detections": detections}
        except Exception as e:
            logger.error("Failed to analyze candles: %s", str(e))
            return {"prompt": prompt, "description": "Error analyzing candles. Try a clearer screenshot with visible candles and prices."}

    else:
        return {"prompt": prompt, "description": "Unsupported prompt. Use 'Was siehst du auf dem Bild?' or 'List last 8 candles with their colors'."