import tkinter as tk from tkinter import filedialog, messagebox from PIL import Image, ImageTk import cv2 import numpy as np from openvino import Core import os import pickle import time ie = Core() devices = ie.available_devices device_to_use = "NPU" if "NPU" in devices else "CPU" print(f"Folosesc inițial: {device_to_use}") model_folder = "models" if not (os.path.exists(os.path.join(model_folder, "face-detection-adas-0001.xml")) and os.path.exists(os.path.join(model_folder, "face-detection-adas-0001.bin")) and os.path.exists(os.path.join(model_folder, "face-reidentification-retail-0095.xml")) and os.path.exists(os.path.join(model_folder, "face-reidentification-retail-0095.bin"))): print("Modelele lipsesc. Descarcă manual din OpenVINO Model Zoo și pune-le în folderul 'models'.") exit() try: model_det = ie.read_model(model=os.path.join(model_folder, "face-detection-adas-0001.xml")) model_rec = ie.read_model(model=os.path.join(model_folder, "face-reidentification-retail-0095.xml")) compiled_det = ie.compile_model(model_det, device_to_use) compiled_rec = ie.compile_model(model_rec, device_to_use) except Exception as e: print(f"Eroare la încărcarea modelelor: {e}") exit() db_file = "face_database.pkl" database = {} if os.path.exists(db_file): with open(db_file, "rb") as f: database = pickle.load(f) npu_failed = False def switch_to_cpu(): global compiled_det, compiled_rec, device_to_use, npu_failed device_to_use = "CPU" compiled_det = ie.compile_model(model_det, device_to_use) compiled_rec = ie.compile_model(model_rec, device_to_use) npu_failed = True print(f"Comutat pe: {device_to_use} (NPU dezactivat permanent)") def detect_face(image): blob = cv2.dnn.blobFromImage(image, size=(672, 384), ddepth=cv2.CV_8U) try: outputs = compiled_det.infer_new_request({0: blob}) except RuntimeError as e: if not npu_failed: print(f"Eroare NPU: {e}. Comut pe CPU.") switch_to_cpu() outputs = compiled_det.infer_new_request({0: blob}) else: raise boxes = outputs[compiled_det.outputs[0].get_any_name()] for box in boxes[0][0]: if box[2] > 0.5: x_min, y_min, x_max, y_max = (box[3:7] * np.array([image.shape[1], image.shape[0], image.shape[1], image.shape[0]])).astype(int) return image[y_min:y_max, x_min:x_max] return None def get_embedding(face): blob = cv2.dnn.blobFromImage(face, size=(128, 128)) try: embedding = compiled_rec.infer_new_request({0: blob}) except RuntimeError as e: if not npu_failed: print(f"Eroare NPU: {e}. Comut pe CPU.") switch_to_cpu() embedding = compiled_rec.infer_new_request({0: blob}) else: raise return embedding[compiled_rec.outputs[0].get_any_name()].flatten() def compare_faces(emb1, emb2): return np.dot(emb1, emb2) / (np.linalg.norm(emb1) * np.linalg.norm(emb2)) root = tk.Tk() root.title("Bază de Date Fețe") root.geometry("600x400") frame_db = tk.LabelFrame(root, text="Bază de Date") frame_db.pack(side=tk.LEFT, padx=10, pady=10) frame_test = tk.LabelFrame(root, text="Imagine Test") frame_test.pack(side=tk.RIGHT, padx=10, pady=10) label_db = tk.Label(frame_db) label_db.pack() label_test = tk.Label(frame_test) label_test.pack() def add_multiple_to_database(): files = filedialog.askopenfilenames(filetypes=[("Image files", "*.jpg *.png")]) if files: added_count = 0 skipped_count = 0 for file in files: img = cv2.imread(file) if img is None or img.size == 0: skipped_count += 1 continue img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) face = detect_face(img) if face is None or face.size == 0: skipped_count += 1 continue embedding = get_embedding(face) name = os.path.basename(file).split('.')[0] is_duplicate = False for existing_name, existing_emb in database.items(): similarity = compare_faces(embedding, existing_emb) if similarity > 0.9: is_duplicate = True skipped_count += 1 break if not is_duplicate: database[name] = embedding added_count += 1 face_img = Image.fromarray(face).resize((150, 150)) face_img_tk = ImageTk.PhotoImage(face_img) label_db.config(image=face_img_tk) label_db.image = face_img_tk if not npu_failed: # Adaugă întârziere doar pe NPU time.sleep(0.5) # 0.5 secunde = 2 imagini pe secundă with open(db_file, "wb") as f: pickle.dump(database, f) messagebox.showinfo("Succes", f"{added_count} fețe adăugate, {skipped_count} sărite (fără fețe sau duplicate)!") def check_face(): if not database: messagebox.showwarning("Atenție", "Baza de date este goală! Adaugă imagini mai întâi.") return file = filedialog.askopenfilename(filetypes=[("Image files", "*.jpg *.png")]) if file: img = cv2.imread(file) if img is None or img.size == 0: messagebox.showerror("Eroare", "Imaginea nu poate fi încărcată!") return img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) face = detect_face(img) if face is not None and face.size > 0: test_embedding = get_embedding(face) test_img = Image.fromarray(face).resize((150, 150)) test_img_tk = ImageTk.PhotoImage(test_img) label_test.config(image=test_img_tk) label_test.image = test_img_tk best_match = None best_similarity = -1 for name, embedding in database.items(): similarity = compare_faces(test_embedding, embedding) if similarity > best_similarity: best_similarity = similarity best_match = name if best_similarity > 0.8: messagebox.showinfo("Rezultat", f"Persoana este: {best_match}\nSimilaritate: {best_similarity:.2f}") else: messagebox.showinfo("Rezultat", f"Nu s-a găsit o potrivire.\nCea mai apropiată: {best_match} (similaritate: {best_similarity:.2f})") else: messagebox.showerror("Eroare", "Nu s-a detectat față!") btn_add = tk.Button(root, text="Adaugă Imagini", command=add_multiple_to_database) btn_add.pack(pady=5) btn_test = tk.Button(root, text="Verifică Față", command=check_face) btn_test.pack(pady=5) root.mainloop()