from PIL import Image, ImageDraw, ImageFont import numpy as np import gradio as gr import os import cv2 from deepface import DeepFace from nudenet import NudeDetector import io import tempfile # --- Constants for image generation --- FONT_SIZE_TITLE = 36 # Increased from 28 FONT_SIZE_DESC = 24 # Increased from 20 PADDING = 30 ITEM_SPACING = 50 BORDER_COLOR = (255, 20, 147) BORDER_WIDTH = 3 TEXT_COLOR = (255, 20, 147) BACKGROUND_COLOR = (0, 0, 0) ERROR_IMAGE_WIDTH = 800 ERROR_IMAGE_HEIGHT = 400 TARGET_MAX_DIMENSION = 400 PIXELS_PER_CM_ESTIMATE = 15 MAX_PROCESSING_DIMENSION = 1024 # New constant for initial image resizing # Try to load a font, fall back to default if not available try: FONT_TITLE = ImageFont.truetype("arial.ttf", FONT_SIZE_TITLE) try: FONT_DESC = ImageFont.truetype("arialbd.ttf", FONT_SIZE_DESC) except IOError: print("Could not load arialbd.ttf, falling back to arial.ttf for FONT_DESC.") FONT_DESC = ImageFont.truetype("arial.ttf", FONT_SIZE_DESC) FONT_ERROR = ImageFont.truetype("arial.ttf", 30) # Increased from 24 FONT_LABEL_SIMPLE = ImageFont.truetype("arial.ttf", 18) # Increased from 14 except IOError: print("Could not load arial.ttf, using default font.") FONT_TITLE = ImageFont.load_default() FONT_DESC = ImageFont.load_default() FONT_ERROR = ImageFont.load_default() FONT_LABEL_SIMPLE = ImageFont.load_default() def create_error_image(message): error_img = Image.new('RGB', (ERROR_IMAGE_WIDTH, ERROR_IMAGE_HEIGHT), color=BACKGROUND_COLOR) draw = ImageDraw.Draw(error_img) bbox = FONT_ERROR.getbbox(message) text_width = bbox[2] - bbox[0] text_height = bbox[3] - bbox[1] x = (ERROR_IMAGE_WIDTH - text_width) / 2 y = (ERROR_IMAGE_HEIGHT - text_height) / 2 draw.text((x, y), message, font=FONT_ERROR, fill=TEXT_COLOR) return error_img def resize_crop(crop_img, target_max_dim): if crop_img.width == 0 or crop_img.height == 0: return crop_img if crop_img.width > crop_img.height: new_width = target_max_dim new_height = int(crop_img.height * (target_max_dim / crop_img.width)) else: new_height = target_max_dim new_width = int(crop_img.width * (target_max_dim / crop_img.height)) return crop_img.resize((new_width, new_height), Image.Resampling.LANCZOS) def estimate_breast_size_cm(crop_img): width_px, height_px = crop_img.size width_cm = round(width_px / PIXELS_PER_CM_ESTIMATE, 1) height_cm = round(height_px / PIXELS_PER_CM_ESTIMATE, 1) return f"Breite: {width_cm} cm · Höhe: {height_cm} cm (geschätzt)" def estimate_vagina_size_cm(crop_img): width_px, height_px = crop_img.size width_cm = round(width_px / PIXELS_PER_CM_ESTIMATE, 1) height_cm = round(height_px / PIXELS_PER_CM_ESTIMATE, 1) return f"Breite: {width_cm} cm · Höhe: {height_cm} cm (geschätzt)" def estimate_body_measurements(person_bbox, associated_nudenet_dets, analysis_mode): pb_x1, pb_y1, pb_x2, pb_y2 = person_bbox person_width_px = pb_x2 - pb_x1 person_height_px = pb_y2 - pb_y1 if person_width_px <= 0 or person_height_px <= 0: return "Körpermaße nicht schätzbar (ungültiger Personen-BBox)" person_width_cm = person_width_px / PIXELS_PER_CM_ESTIMATE person_height_cm = person_height_px / PIXELS_PER_CM_ESTIMATE bust_cm = "?" waist_cm = "?" hip_cm = "?" breast_detections = [] vagina_detections = [] if analysis_mode in ['Komplett (Brüste + Vagina + geschätztes Alter)', 'Brüste (Nur Brüste)']: breast_detections = [det for det in associated_nudenet_dets if 'FEMALE_BREAST_EXPOSED' in det['class']] if analysis_mode in ['Komplett (Brüste + Vagina + geschätztes Alter)', 'Vagina (Nur Vagina)']: vagina_detections = [det for det in associated_nudenet_dets if 'FEMALE_GENITALIA_EXPOSED' in det['class']] if breast_detections: # Simple estimate: average breast width b_width_px = sum([det['box'][2] for det in breast_detections]) / len(breast_detections) bust_cm = round(b_width_px * 2.5 / PIXELS_PER_CM_ESTIMATE, 1) # A rough multiplier for bust circumference if vagina_detections: vg_width_px = max([det['box'][2] for det in vagina_detections]) hip_cm = round(vg_width_px * 2.0 / PIXELS_PER_CM_ESTIMATE, 1) elif person_height_px > 0: hip_cm = round(person_width_cm * 0.9, 1) if person_width_cm > 0: waist_cm = round(person_width_cm * 0.8, 1) return f"Maße (geschätzt, unsicher): Büste: {bust_cm} cm · Taille: {waist_cm} cm · Hüfte: {hip_cm} cm" def describe_breast_precise(crop): w, h = crop.size crop_np = np.array(crop) gray = cv2.cvtColor(crop_np, cv2.COLOR_RGB2GRAY) # Calculate nipple area/prominence (very rough estimation) # This is a very rough estimation and needs more advanced computer vision # For now, let's just make a simple placeholder based on contrast/shape # Find contours, look for circular/elliptical shapes _, thresh = cv2.threshold(gray, 100, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU) contours, _ = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE) nipple_detected = False for contour in contours: area = cv2.contourArea(contour) if area > 50 and area < (w * h / 5): # Filter by reasonable size perimeter = cv2.arcLength(contour, True) if perimeter > 0: circularity = 4 * np.pi * area / (perimeter * perimeter) if circularity > 0.6: # Check for somewhat circular shapes nipple_detected = True break nipple_prominence = "Sichtbar" if nipple_detected else "Nicht hervorstehend" # Estimate shape (round, conical, bell, etc.) ratio_wh = w / h if ratio_wh > 1.1: shape = "Breit / Horizontal" elif ratio_wh < 0.9: shape = "Hoch / Vertikal" else: shape = "Rund / Ausgewogen" size = "klein" if w*h < 30000 else "mittel" if w*h < 80000 else "groß" if w*h < 150000 else "sehr groß" size_cm_estimate = estimate_breast_size_cm(crop) return f"Form: {shape} · Größe: {size} · Nippel: {nipple_prominence} · Maße: {size_cm_estimate}" def describe_vagina_precise(crop): w, h = crop.size crop_np = np.array(crop) gray = cv2.cvtColor(crop_np, cv2.COLOR_RGB2GRAY) hair_mask = cv2.inRange(gray, 40, 140) hair_ratio = np.sum(hair_mask > 0) / (w*h) if hair_ratio < 0.03: shaved = "komplett glatt rasiert" elif hair_ratio < 0.12: shaved = "minimal (Landing Strip / Dreieck)" elif hair_ratio < 0.35: shaved = "teilrasiert / Brazilian" else: shaved = "voll behaart (Bush)" ratio_wh = w / h labia_area = w * h # More detailed form descriptions if labia_area < 20000: # Very small overall area if ratio_wh > 1.0: form = "Mini Outie (Schmetterlingsform)" else: form = "Sehr kleine Innie (Barbie-Form)" elif ratio_wh < 0.75: # Taller than wide form = "Lange Innie (Vertikale Form)" elif ratio_wh > 1.5: # Very wide if labia_area > 70000: form = "Extrem ausgeprägtes Outie (Puff-Form)" else: form = "Breites Outie (Offene Form)" elif ratio_wh > 1.1: # Moderately wide (more typical Outie) form = "Klassisches Outie (Labien sichtbar)" else: # Ratio between 0.75 and 1.1, and not tiny area (more typical Innie) form = "Klassische Innie (Geschlossene Form)" size = "winzig" if labia_area < 20000 else "klein" if labia_area < 40000 else "mittel" if labia_area < 70000 else "groß & voll" size_cm_estimate = estimate_vagina_size_cm(crop) return f"Form: {form} · Größe: {size} · Behaart: {shaved} · Maße: {size_cm_estimate}" def estimate_age(crop_img): try: img_array = np.array(crop_img) face = DeepFace.extract_faces(img_array, detector_backend='retinaface', enforce_detection=False) if len(face) > 0: age = DeepFace.analyze(img_array, actions=['age'], enforce_detection=False, silent=True)[0]['age'] return f"ca. {int(age)} Jahre (±3)" else: gray = cv2.cvtColor(img_array, cv2.COLOR_RGB2GRAY) smoothness = cv2.Laplacian(gray, cv2.CV_64F).var() if smoothness > 1800: return "jung (18–25)" elif smoothness > 1000: return "jung-mittel (25–35)" else: return "reif (35–45+)" except Exception as e: print(f"Error estimating age: {e}") return "Alter nicht bestimmbar" # Initialize NudeDetector detector = NudeDetector() def _analyze_single_image_and_generate_outputs(analysis_mode, input_image_path): print(f"[INFO] _analyze_single_image_and_generate_outputs called with mode: {analysis_mode}, path: {input_image_path}") temp_img_path = None try: if input_image_path is None: print("[WARNING] No input image path provided to single analyzer.") return create_error_image('Kein Bild ausgewählt.'), None processing_img_for_detection_path = input_image_path try: original_img = Image.open(input_image_path).convert("RGB") if max(original_img.width, original_img.height) > MAX_PROCESSING_DIMENSION: original_img = resize_crop(original_img, MAX_PROCESSING_DIMENSION) print(f"[INFO] Image resized to {original_img.width}x{original_img.height} for faster processing.") with tempfile.NamedTemporaryFile(suffix=".jpg", delete=False) as tmp_file: original_img.save(tmp_file.name, "JPEG") temp_img_path = tmp_file.name processing_img_for_detection_path = temp_img_path else: processing_img_for_detection_path = input_image_path original_img_np = np.array(original_img) print(f"[INFO] Image opened and potentially resized successfully: {input_image_path}") except Exception as e: print(f"[ERROR] Failed to open image {input_image_path}: {e}") return create_error_image(f'Fehler beim Öffnen des Bildes: {e}'), None current_filename = os.path.basename(input_image_path) try: nudenet_detections = detector.detect(processing_img_for_detection_path) # Filter NudeNet detections based on analysis_mode allowed_classes = [] if analysis_mode == 'Komplett (Brüste + Vagina + geschätztes Alter)': allowed_classes = ['FEMALE_GENITALIA_EXPOSED', 'ANUS_EXPOSED', 'FEMALE_BREAST_EXPOSED', 'BUTTOCKS_EXPOSED'] elif analysis_mode == 'Brüste (Nur Brüste)': allowed_classes = ['FEMALE_BREAST_EXPOSED', 'ANUS_EXPOSED', 'BUTTOCKS_EXPOSED'] # Include other body parts for general detection context elif analysis_mode == 'Vagina (Nur Vagina)': allowed_classes = ['FEMALE_GENITALIA_EXPOSED', 'ANUS_EXPOSED', 'BUTTOCKS_EXPOSED'] # Include other body parts for general detection context if allowed_classes: nudenet_detections = [det for det in nudenet_detections if det['class'] in allowed_classes] print(f"[INFO] NudeNet detected {len(nudenet_detections)} objects for mode '{analysis_mode}'.") except Exception as e: print(f"[ERROR] Error during NudeNet detection for {input_image_path}: {e}") return create_error_image(f'Fehler bei der NudeNet-Erkennung: {e}'), None persons_data = [] face_detections = [] # Face detection for age estimation is always attempted if mode includes age if analysis_mode == 'Komplett (Brüste + Vagina + geschätztes Alter)': try: face_results = DeepFace.extract_faces(original_img_np, detector_backend='retinaface', enforce_detection=False) face_detections = [f for f in face_results if f['confidence'] > 0.9] print(f"[INFO] DeepFace detected {len(face_detections)} faces.") except Exception as e: print(f"[WARNING] Error during DeepFace face detection: {e}. Proceeding without person association or age estimation.") if face_detections: for face_det in face_detections: x, y, w, h = face_det['facial_area']['x'], face_det['facial_area']['y'], face_det['facial_area']['w'], face_det['facial_area']['h'] expand_factor_y = 2.0 expand_factor_x = 1.2 person_center_x = x + w / 2 person_center_y = y + h / 2 new_h = int(h * expand_factor_y) new_w = int(w * expand_factor_x) person_x1 = max(0, int(person_center_x - new_w / 2)) person_y1 = max(0, int(person_center_y - new_h / 2)) person_x2 = min(original_img.width, int(person_center_x + new_w / 2)) person_y2 = min(original_img.height, int(person_center_y + new_h / 2)) person_bbox = (person_x1, person_y1, person_x2, person_y2) associated_nudenet_dets = [] for nude_det in nudenet_detections: nx1, ny1, nx2, ny2 = nude_det['box'][0], nude_det['box'][1], nude_det['box'][0] + nude_det['box'][2], nude_det['box'][1] + nude_det['box'][3] nude_center_x = nx1 + (nx2 - nx1) / 2 nude_center_y = ny1 + (ny2 - ny1) / 2 if person_x1 <= nude_center_x <= person_x2 and person_y1 <= nude_center_y <= person_y2: associated_nudenet_dets.append(nude_det) if associated_nudenet_dets: body_measurements_str = estimate_body_measurements(person_bbox, associated_nudenet_dets, analysis_mode) persons_data.append({ 'person_bbox': person_bbox, 'nudenet_detections': associated_nudenet_dets, 'face_area': face_det['facial_area'], 'body_measurements': body_measurements_str }) print(f"[INFO] {len(persons_data)} persons with associated NudeNet detections found for mode '{analysis_mode}'.") annotated_original_img = original_img.copy() draw_annotated = ImageDraw.Draw(annotated_original_img) current_image_crops_with_details = [] processed_nudenet_ids = set() for i, person in enumerate(persons_data): px1, py1, px2, py2 = person['person_bbox'] # Draw person bounding box if there are associated detections if any(det['class'] in allowed_classes for det in person['nudenet_detections']): draw_annotated.rectangle([(px1, py1), (px2, py2)], outline=(0, 255, 0), width=BORDER_WIDTH) label_person = f"Person {i+1}" bbox_label_person = FONT_LABEL_SIMPLE.getbbox(label_person) text_width_person = bbox_label_person[2] - bbox_label_person[0] text_height_person = bbox_label_person[3] - bbox_label_person[1] draw_annotated.rectangle([(px1, py1 - text_height_person - 2), (px1 + text_width_person, py1)], fill=(0, 255, 0)) draw_annotated.text((px1, py1 - text_height_person - 2), label_person, font=FONT_LABEL_SIMPLE, fill=BACKGROUND_COLOR) if person['body_measurements'] and person['body_measurements'] != "Körpermaße nicht schätzbar (ungültiger Personen-BBox)": current_image_crops_with_details.append((f"Person {i+1} Körpermaße", Image.new('RGB', (1,1), (0,0,0)), person['body_measurements'])) for det in person['nudenet_detections']: det_class = det['class'] x1, y1, x2, y2 = det['box'][0], det['box'][1], det['box'][0] + det['box'][2], det['box'][1] + det['box'][3] x1, y1, x2, y2 = max(0, x1), max(0, y1), min(original_img.width, x2), min(original_img.height, y2) draw_annotated.rectangle([(x1, y1), (x2, y2)], outline=BORDER_COLOR, width=BORDER_WIDTH) crop = original_img.crop((x1, y1, x2, y2)) age_str = 'Alter unbekannt' if analysis_mode == 'Komplett (Brüste + Vagina + geschätztes Alter)': try: face_x, face_y, face_w, face_h = person['face_area']['x'], person['face_area']['y'], person['face_area']['w'], person['face_area']['h'] face_crop_for_age = original_img.crop((face_x, face_y, face_x + face_w, face_y + face_h)) age_str = estimate_age(face_crop_for_age) except Exception as e: print(f"[ERROR] Error estimating age for person {i+1} crop: {e}") if 'FEMALE_GENITALIA_EXPOSED' in det_class and analysis_mode in ['Komplett (Brüste + Vagina + geschätztes Alter)', 'Vagina (Nur Vagina)']: label_detailed = f"Person {i+1} Vagina" desc = describe_vagina_precise(crop) if analysis_mode == 'Komplett (Brüste + Vagina + geschätztes Alter)': desc += f" · Alter: {age_str}" current_image_crops_with_details.append((label_detailed, resize_crop(crop, TARGET_MAX_DIMENSION), desc)) processed_nudenet_ids.add(id(det)) elif 'FEMALE_BREAST_EXPOSED' in det_class and analysis_mode in ['Komplett (Brüste + Vagina + geschätztes Alter)', 'Brüste (Nur Brüste)']: label_detailed = f"Person {i+1} Brust" desc = describe_breast_precise(crop) if analysis_mode == 'Komplett (Brüste + Vagina + geschätztes Alter)': desc += f" · Alter: {age_str}" current_image_crops_with_details.append((label_detailed, resize_crop(crop, TARGET_MAX_DIMENSION), desc)) processed_nudenet_ids.add(id(det)) elif ('ANUS_EXPOSED' in det_class or 'BUTTOCKS_EXPOSED' in det_class) and analysis_mode == 'Komplett (Brüste + Vagina + geschätztes Alter)': label_detailed = f"Person {i+1} {det_class.replace('_EXPOSED', '').replace('_', ' ').title()}" desc = f"Erkannt: {det_class.replace('_EXPOSED', '').replace('_', ' ').title()}" if analysis_mode == 'Komplett (Brüste + Vagina + geschätztes Alter)': desc += f" · Alter: {age_str}" current_image_crops_with_details.append((label_detailed, resize_crop(crop, TARGET_MAX_DIMENSION), desc)) processed_nudenet_ids.add(id(det)) unassociated_nudenet_detections = [det for det in nudenet_detections if id(det) not in processed_nudenet_ids] for det in unassociated_nudenet_detections: det_class = det['class'] x1, y1, x2, y2 = det['box'][0], det['box'][1], det['box'][0] + det['box'][2], det['box'][1] + det['box'][3] x1, y1, x2, y2 = max(0, x1), max(0, y1), min(original_img.width, x2), min(original_img.height, y2) draw_annotated.rectangle([(x1, y1), (x2, y2)], outline=(255, 255, 0), width=BORDER_WIDTH) label_global_simple = f"Global {det_class.replace('_EXPOSED', '').replace('_', ' ').title()}" bbox_label_global = FONT_LABEL_SIMPLE.getbbox(label_global_simple) text_width_global = bbox_label_global[2] - bbox_label_global[0] text_height_global = bbox_label_global[3] - bbox_label_global[1] draw_annotated.rectangle([(x1, y1 - text_height_global - 2), (x1 + text_width_global, y1)], fill=TEXT_COLOR) draw_annotated.text((x1, y1 - text_height_global - 2), label_global_simple, font=FONT_LABEL_SIMPLE, fill=BACKGROUND_COLOR) crop = original_img.crop((x1, y1, x2, y2)) age_str = 'Alter unbekannt' if analysis_mode == 'Komplett (Brüste + Vagina + geschätztes Alter)': try: resized_crop_for_age = resize_crop(crop, TARGET_MAX_DIMENSION) age_str = estimate_age(resized_crop_for_age) except Exception as e: print(f"[ERROR] Error estimating age for global crop: {e}") if 'FEMALE_GENITALIA_EXPOSED' in det_class and analysis_mode in ['Komplett (Brüste + Vagina + geschätztes Alter)', 'Vagina (Nur Vagina)']: label_detailed = "Global Vagina" desc = describe_vagina_precise(crop) if analysis_mode == 'Komplett (Brüste + Vagina + geschätztes Alter)': desc += f" · Alter: {age_str}" current_image_crops_with_details.append((label_detailed, resize_crop(crop, TARGET_MAX_DIMENSION), desc)) elif 'FEMALE_BREAST_EXPOSED' in det_class and analysis_mode in ['Komplett (Brüste + Vagina + geschätztes Alter)', 'Brüste (Nur Brüste)']: label_detailed = "Global Brust" desc = describe_breast_precise(crop) if analysis_mode == 'Komplett (Brüste + Vagina + geschätztes Alter)': desc += f" · Alter: {age_str}" current_image_crops_with_details.append((label_detailed, resize_crop(crop, TARGET_MAX_DIMENSION), desc)) elif ('ANUS_EXPOSED' in det_class or 'BUTTOCKS_EXPOSED' in det_class) and analysis_mode == 'Komplett (Brüste + Vagina + geschätztes Alter)': label_detailed = f"Global {det_class.replace('_EXPOSED', '').replace('_', ' ').title()}" desc = f"Erkannt: {det_class.replace('_EXPOSED', '').replace('_', ' ').title()}" if analysis_mode == 'Komplett (Brüste + Vagina + geschätztes Alter)': desc += f" · Alter: {age_str}" current_image_crops_with_details.append((label_detailed, resize_crop(crop, TARGET_MAX_DIMENSION), desc)) if not current_image_crops_with_details: text = "Keine relevanten Bereiche für den ausgewählten Analysemodus erkannt." original_img_width, original_img_height = original_img.size bbox_text = FONT_TITLE.getbbox(text) text_width = bbox_text[2] - bbox_text[0] text_height = bbox_text[3] - bbox_text[1] draw_annotated.text(((original_img_width - text_width) / 2, original_img_height - text_height - PADDING), text, font=FONT_TITLE, fill=TEXT_COLOR) print(f"[INFO] No relevant detections for mode '{analysis_mode}', adding message to original image.") if not current_image_crops_with_details: composite_crops_img = create_error_image("Keine detailreichen Bereiche für Crops gefunden.") else: max_crop_width = 0 total_height_crops = PADDING for title, crop_img, desc in current_image_crops_with_details: if crop_img.width == 1 and crop_img.height == 1: # This is for body measurements bbox_title = FONT_TITLE.getbbox(title) bbox_desc = FONT_DESC.getbbox(desc) text_title_height = bbox_title[3] - bbox_title[1] text_desc_height = bbox_desc[3] - bbox_desc[1] total_height_crops += text_title_height + text_desc_height + PADDING * 2 + ITEM_SPACING max_crop_width = max(max_crop_width, (bbox_title[2] - bbox_title[0]), (bbox_desc[2] - bbox_desc[0])) else: max_crop_width = max(max_crop_width, crop_img.width) bbox_title = FONT_TITLE.getbbox(title) bbox_desc = FONT_DESC.getbbox(desc) text_title_height = bbox_title[3] - bbox_title[1] text_desc_height = bbox_desc[3] - bbox_desc[1] total_height_crops += crop_img.height + text_title_height + text_desc_height + PADDING * 2 + ITEM_SPACING total_height_crops += PADDING canvas_width_crops = max_crop_width + PADDING * 2 for title, crop_img, desc in current_image_crops_with_details: bbox_title = FONT_TITLE.getbbox(title) bbox_desc = FONT_DESC.getbbox(desc) text_title_width = bbox_title[2] - bbox_title[0] text_desc_width = bbox_desc[2] - bbox_desc[0] canvas_width_crops = max(canvas_width_crops, text_title_width + PADDING * 2, text_desc_width + PADDING * 2) composite_crops_img = Image.new('RGB', (canvas_width_crops, total_height_crops), color = BACKGROUND_COLOR) draw_crops = ImageDraw.Draw(composite_crops_img) current_y_offset_crops = PADDING for title, crop_img, desc in current_image_crops_with_details: if crop_img.width == 1 and crop_img.height == 1: # This is for body measurements bbox_title = FONT_TITLE.getbbox(title) text_title_width = bbox_title[2] - bbox_title[0] text_title_height = bbox_title[3] - bbox_title[1] draw_crops.text(((canvas_width_crops - text_title_width) / 2, current_y_offset_crops), title, font=FONT_TITLE, fill=TEXT_COLOR) current_y_offset_crops += text_title_height + PADDING // 2 bbox_desc = FONT_DESC.getbbox(desc) text_desc_width = bbox_desc[2] - bbox_desc[0] text_desc_height = bbox_desc[3] - bbox_desc[1] draw_crops.text(((canvas_width_crops - text_desc_width) / 2, current_y_offset_crops), desc, font=FONT_DESC, fill=TEXT_COLOR) current_y_offset_crops += text_desc_height + ITEM_SPACING else: x_offset_crop = (canvas_width_crops - crop_img.width) // 2 draw_crops.rectangle( (x_offset_crop - BORDER_WIDTH, current_y_offset_crops - BORDER_WIDTH, x_offset_crop + crop_img.width + BORDER_WIDTH, current_y_offset_crops + crop_img.height + BORDER_WIDTH), fill=BORDER_COLOR ) composite_crops_img.paste(crop_img, (x_offset_crop, current_y_offset_crops)) current_y_offset_crops += crop_img.height + PADDING bbox_title = FONT_TITLE.getbbox(title) text_title_width = bbox_title[2] - bbox_title[0] text_title_height = bbox_title[3] - bbox_title[1] draw_crops.text(((canvas_width_crops - text_title_width) / 2, current_y_offset_crops), title, font=FONT_TITLE, fill=TEXT_COLOR) current_y_offset_crops += text_title_height + PADDING // 2 bbox_desc = FONT_DESC.getbbox(desc) text_desc_width = bbox_desc[2] - bbox_desc[0] text_desc_height = bbox_desc[3] - bbox_desc[1] draw_crops.text(((canvas_width_crops - text_desc_width) / 2, current_y_offset_crops), desc, font=FONT_DESC, fill=TEXT_COLOR) current_y_offset_crops += text_desc_height + ITEM_SPACING separator_text_original = "Originalbild mit Markierungen" separator_text_crops = "Detaillierte Analysen" if analysis_mode == 'Brüste (Nur Brüste)': separator_text_crops = "Detaillierte Brust-Analysen" elif analysis_mode == 'Vagina (Nur Vagina)': separator_text_crops = "Detaillierte Vagina-Analysen" elif analysis_mode == 'Komplett (Brüste + Vagina + geschätztes Alter)': separator_text_crops = "Detaillierte Gesamt-Analysen" bbox_separator_original = FONT_TITLE.getbbox(separator_text_original) bbox_separator_crops = FONT_TITLE.getbbox(separator_text_crops) separator_height_total = (bbox_separator_original[3] - bbox_separator_original[1]) + (bbox_separator_crops[3] - bbox_separator_crops[1]) + PADDING * 4 combined_width = max(annotated_original_img.width, composite_crops_img.width, (bbox_separator_original[2] - bbox_separator_original[0]) + PADDING * 2, (bbox_separator_crops[2] - bbox_separator_crops[0]) + PADDING * 2) combined_width = max(combined_width, ERROR_IMAGE_WIDTH) combined_height = annotated_original_img.height + separator_height_total + composite_crops_img.height + PADDING * 2 combined_img = Image.new('RGB', (combined_width, combined_height), color=BACKGROUND_COLOR) combined_draw = ImageDraw.Draw(combined_img) current_y_combined = PADDING text_width_orig_sep = bbox_separator_original[2] - bbox_separator_original[0] text_height_orig_sep = bbox_separator_original[3] - bbox_separator_original[1] combined_draw.text(((combined_width - text_width_orig_sep) / 2, current_y_combined), separator_text_original, font=FONT_TITLE, fill=TEXT_COLOR) current_y_combined += text_height_orig_sep + PADDING x_offset_original = (combined_width - annotated_original_img.width) // 2 combined_img.paste(annotated_original_img, (x_offset_original, current_y_combined)) current_y_combined += annotated_original_img.height + PADDING text_width_crops_sep = bbox_separator_crops[2] - bbox_separator_crops[0] text_height_crops_sep = bbox_separator_crops[3] - bbox_separator_crops[1] combined_draw.text(((combined_width - text_width_crops_sep) / 2, current_y_combined), separator_text_crops, font=FONT_TITLE, fill=TEXT_COLOR) current_y_combined += text_height_crops_sep + PADDING x_offset_crops = (combined_width - composite_crops_img.width) // 2 combined_img.paste(composite_crops_img, (x_offset_crops, current_y_combined)) print("[INFO] Combined image generated successfully.") with tempfile.NamedTemporaryFile(suffix=".pdf", delete=False) as tmp_pdf: combined_img.save(tmp_pdf.name, "PDF") pdf_path = tmp_pdf.name return combined_img, pdf_path except Exception as e: print(f"[CRITICAL ERROR] An unexpected error occurred during single image analysis: {e}") if temp_img_path and os.path.exists(temp_img_path): os.remove(temp_img_path) return create_error_image(f'Unerwarteter Fehler bei der Bildanalyse für {os.path.basename(input_image_path)}: {e}'), None finally: if temp_img_path and os.path.exists(temp_img_path): os.remove(temp_img_path) def analyze_images_for_gallery(analysis_mode, input_image_paths): print(f"[INFO] analyze_images_for_gallery called with mode: {analysis_mode} and {len(input_image_paths) if input_image_paths else 0} paths.") all_combined_images = [] all_pdf_paths = [] if not input_image_paths: return [create_error_image('Keine Bilder ausgewählt.')], [None] for input_image_path in input_image_paths: combined_img, pdf_path = _analyze_single_image_and_generate_outputs(analysis_mode, input_image_path) all_combined_images.append(combined_img) all_pdf_paths.append(pdf_path) return all_combined_images, all_pdf_paths interface = gr.Interface( fn=analyze_images_for_gallery, inputs=[ gr.Radio( choices=['Komplett (Brüste + Vagina + geschätztes Alter)', 'Brüste (Nur Brüste)', 'Vagina (Nur Vagina)'], value='Komplett (Brüste + Vagina + geschätztes Alter)', label='Analyse Modus' ), gr.File(file_count="multiple", type="filepath", label="Bilder hochladen (mehrere Dateien erlaubt)") ], outputs=[ gr.Gallery(label="Analysierte Kompositbilder"), gr.File(file_count="multiple", label="PDFs herunterladen") ], title="🔞 Nacktheits-Analysator", description="Laden Sie ein oder mehrere Bilder hoch, um eine detaillierte Analyse basierend auf dem ausgewählten Modus zu erhalten. Das Ergebnis ist ein Kompositbild mit Originalbild und detaillierten Crops, pro Person gruppiert. Sie können auch eine PDF-Version herunterladen." ) if __name__ == "__main__": interface.launch(debug=True, share=True)