🔍 نظام تحليل الصور المتطور

from flask import Flask, request, render_template_string from PIL import Image, ImageEnhance from facenet_pytorch import MTCNN from transformers import AutoImageProcessor, AutoModelForImageClassification import torch from torch.nn.functional import softmax import numpy as np import cv2 import warnings import base64 import io import os warnings.filterwarnings('ignore') app = Flask(__name__) app.config['MAX_CONTENT_LENGTH'] = 10 * 1024 * 1024 device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') print(f"🔧 جهاز التشغيل: {device}") print("⏳ جاري تحميل النماذج...") # كشف الوجه mtcnn = MTCNN(keep_all=True, device=device, min_face_size=40, thresholds=[0.6, 0.7, 0.7]) print("✅ MTCNN") # نموذج ViT للجنس والعمر try: vit_processor = AutoImageProcessor.from_pretrained("abhilash88/age-gender-prediction") vit_model = AutoModelForImageClassification.from_pretrained("abhilash88/age-gender-prediction").to(device) vit_model.eval() print("✅ ViT Age-Gender (94.3%)") except Exception as e: vit_model = None print(f"⚠️ ViT: {e}") # نموذج MetaCLIP-2 try: clip_processor = AutoImageProcessor.from_pretrained("prithivMLmods/MetaCLIP-2-Gender-Identifier") clip_model = AutoModelForImageClassification.from_pretrained("prithivMLmods/MetaCLIP-2-Gender-Identifier").to(device) clip_model.eval() print("✅ MetaCLIP-2 (97.25%)") except Exception as e: clip_model = None print(f"⚠️ MetaCLIP-2: {e}") # نموذج SigLIP2 لفحص المحتوى try: nsfw_processor = AutoImageProcessor.from_pretrained("prithivMLmods/siglip2-mini-explicit-content") nsfw_model = AutoModelForImageClassification.from_pretrained("prithivMLmods/siglip2-mini-explicit-content").to(device) nsfw_model.eval() print("✅ SigLIP2 NSFW (F1: 0.94)") except Exception as e: nsfw_model = None print(f"⚠️ SigLIP2: {e}") print("✅ جميع النماذج جاهزة\n") def enhance_image(img): try: for cls, factor in [(ImageEnhance.Contrast, 1.2), (ImageEnhance.Sharpness, 1.3), (ImageEnhance.Brightness, 1.05)]: img = cls(img).enhance(factor) return img except: return img def img_to_base64(img): buf = io.BytesIO() img.save(buf, format="JPEG", quality=85) return base64.b64encode(buf.getvalue()).decode() HTML_TEMPLATE = ''' 🔍 نظام تحليل الصور المتطور

{% if error_msg %}

❌ {{ error_msg }}

{% endif %} {% if warning_msg %}

⚠️ {{ warning_msg }}

{% endif %} {% if uploaded_image %}

📷 الصورة

{% endif %} {% if results %}

{% if results.vit %}

🧠 ViT Transformer دقة 94.3%

👤 الجنس: {{ results.vit.gender }}

📊 الثقة: {{ results.vit.confidence }}%

{% endif %} {% if results.clip %}

⚡ MetaCLIP-2 دقة 97.25%

👤 الجنس: {{ results.clip.gender }}

📊 الثقة: {{ results.clip.confidence }}%

{% endif %} {% if results.nsfw %}

🛡 فحص المحتوى F1: 0.94

SigLIP2: {{ results.nsfw.status }}

📊 التفاصيل: {{ results.nsfw.detail }}

{% endif %}

''' def process_image(img): """تنفيذ جميع التحليلات""" results = {"faces": 0, "vit": None, "clip": None, "nsfw": None} # تحسين الصورة img_enhanced = enhance_image(img) # 1. فحص NSFW if nsfw_model: try: inputs = nsfw_processor(images=img_enhanced, return_tensors="pt").to(device) with torch.no_grad(): outputs = nsfw_model(**inputs) probs = softmax(outputs.logits, dim=1).squeeze() safe_prob = probs[4].item() if probs.shape[0] > 4 else probs[0].item() explicit_prob = 1 - safe_prob if explicit_prob > 0.7: status = "⚠️ غير آمن" elif explicit_prob > 0.3: status = "⚠️ مشبوه" else: status = "✅ آمن" results["nsfw"] = { "status": status, "detail": f"نسبة الأمان: {safe_prob*100:.1f}%" } except Exception as e: print(f"خطأ NSFW: {e}") # 2. كشف الوجوه boxes, _ = mtcnn.detect(img_enhanced) if boxes is not None and len(boxes) > 0: results["faces"] = len(boxes) # 3. تحليل ViT if vit_model: try: inputs = vit_processor(images=img_enhanced, return_tensors="pt").to(device) with torch.no_grad(): outputs = vit_model(**inputs) probs = softmax(outputs.logits, dim=1).squeeze() female_prob = probs[0].item() male_prob = probs[1].item() if probs.shape[0] > 1 else 1 - female_prob if male_prob > female_prob: gender = "👨 ذكر (Male)" conf = male_prob * 100 else: gender = "👩 أنثى (Female)" conf = female_prob * 100 results["vit"] = {"gender": gender, "confidence": f"{conf:.1f}"} except Exception as e: print(f"خطأ ViT: {e}") # 4. تحليل MetaCLIP-2 if clip_model: try: inputs = clip_processor(images=img_enhanced, return_tensors="pt").to(device) with torch.no_grad(): outputs = clip_model(**inputs) probs = softmax(outputs.logits, dim=1).squeeze() female_prob = probs[0].item() male_prob = probs[1].item() if probs.shape[0] > 1 else 1 - female_prob if male_prob > female_prob: gender = "👨 ذكر (Male)" conf = male_prob * 100 else: gender = "👩 أنثى (Female)" conf = female_prob * 100 results["clip"] = {"gender": gender, "confidence": f"{conf:.1f}"} except Exception as e: print(f"خطأ MetaCLIP-2: {e}") return results @app.route("/", methods=["GET", "POST"]) def index(): context = {} if request.method == "POST": file = request.files.get("image") if not file or file.filename == "": context["error_msg"] = "❌ الرجاء اختيار ملف صورة" return render_template_string(HTML_TEMPLATE, **context) try: img = Image.open(file).convert("RGB") context["uploaded_image"] = img_to_base64(img) context["results"] = process_image(img) except Exception as e: context["error_msg"] = f"❌ خطأ: {str(e)}" return render_template_string(HTML_TEMPLATE, **context) if __name__ == "__main__": app.run(host="0.0.0.0", port=7860, debug=True, threaded=True)