Upload New File

HW 4/Task1/task1.py

+import cv2 as cv
+import numpy as np
+from scipy.spatial.distance import cityblock, euclidean
+from scipy.stats import pearsonr
+
+# Пути к моделям
+FACE_DETECTION_MODEL = "models/face_detection_yunet_2023mar.onnx"
+FACE_RECOGNITION_MODEL = "models/face_recognition_sface_2021dec.onnx"
+
+# Параметры обработки
+SCORE_THRESHOLD = 0.5
+NMS_THRESHOLD = 0.3
+SCALE = 1.0
+TOP_K = 5000
+
+# Пути к файлам
+GROUND_BASE_IMAGES = ["orig.jpg", "smile.jpg", "think.jpg", "flow.jpg"]  # Несколько эталонных изображений
+VIDEO_INPUT = "spb.mp4"
+VIDEO_OUTPUT = "output_video9.mp4"
+
+
+def load_models():
+    """Загружает модели детекции и распознавания лиц."""
+    detector = cv.FaceDetectorYN.create(
+        FACE_DETECTION_MODEL, "", (320, 320),
+        SCORE_THRESHOLD, NMS_THRESHOLD, TOP_K
+    )
+    recognizer = cv.FaceRecognizerSF.create(FACE_RECOGNITION_MODEL, "")
+    return detector, recognizer
+
+
+def get_ground_base_features(detector, recognizer, image_paths):
+    """Извлекает признаки для нескольких эталонных изображений."""
+    features = []
+    for image_path in image_paths:
+        base_image = cv.imread(cv.samples.findFile(image_path))
+        if base_image is None:
+            print(f"Ошибка загрузки {image_path}")
+            continue
+
+        width, height = base_image.shape[1], base_image.shape[0]
+        detector.setInputSize((width, height))
+        faces = detector.detect(base_image)
+
+        if faces[1] is None:
+            print(f"Лицо не найдено в {image_path}")
+            continue
+
+        face_align = recognizer.alignCrop(base_image, faces[1][0])
+        feature = recognizer.feature(face_align).flatten()
+        features.append(feature)
+
+    assert len(features) > 0, "Не удалось загрузить ни одного эталонного изображения!"
+    return features
+
+
+def compare_faces(recognizer, face_img, reference_features):
+    """Сравнивает лицо с несколькими эталонными изображениями."""
+    feature = recognizer.feature(face_img).flatten()
+
+    metrics_list = []
+    for ref_feature in reference_features:
+        cos_sim = recognizer.match(feature, ref_feature, cv.FaceRecognizerSF_FR_COSINE)
+        l2_sim = recognizer.match(feature, ref_feature, cv.FaceRecognizerSF_FR_NORM_L2)
+        l1_dist = cityblock(feature, ref_feature)
+        euclidean_dist = euclidean(feature, ref_feature)
+        pearson_corr, _ = pearsonr(feature, ref_feature)
+
+        metrics_list.append({
+            "cos_sim": cos_sim,
+            "l2_sim": l2_sim,
+            "l1_dist": l1_dist,
+            "euclidean_dist": euclidean_dist,
+            "pearson_corr": pearson_corr,
+        })
+
+    return metrics_list
+
+
+def determine_face_match(metrics_list):
+    """Определяет, является ли лицо совпадающим, используя несколько эталонных изображений."""
+    for metrics in metrics_list:
+        if (
+            metrics["cos_sim"] >= 0.3 or
+            metrics["l1_dist"] <= 1.0 or
+            metrics["pearson_corr"] > 0.9
+        ):
+            return True
+    return False
+
+
+def process_frame(frame, detector, recognizer, reference_features):
+    """Обрабатывает кадр: обнаружение лиц и рисование рамок."""
+    detector.setInputSize((frame.shape[1], frame.shape[0]))
+    faces = detector.detect(frame)
+
+    if faces[1] is None:
+        return frame
+
+    for face in faces[1]:
+        aligned_face = recognizer.alignCrop(frame, face)
+        metrics_list = compare_faces(recognizer, aligned_face, reference_features)
+        color = (0, 255, 0) if determine_face_match(metrics_list) else (0, 0, 255)
+
+        try:
+            x, y, w, h = face[:4].astype(np.int32)
+        except ValueError:
+            print(f"Ошибка при извлечении координат для лица: {face}")
+            continue
+
+        cv.rectangle(frame, (x, y), (x + w, y + h), color, 2)
+
+    return frame
+
+
+def process_video(input_path, output_path, detector, recognizer, reference_features):
+    """Обрабатывает входное видео и сохраняет результат."""
+    video_capture = cv.VideoCapture(input_path)
+    if not video_capture.isOpened():
+        print("Не удалось открыть видео.")
+        return
+
+    fourcc = cv.VideoWriter_fourcc(*'mp4v')
+    frame_width = int(video_capture.get(cv.CAP_PROP_FRAME_WIDTH) * SCALE)
+    frame_height = int(video_capture.get(cv.CAP_PROP_FRAME_HEIGHT) * SCALE)
+    output_video = cv.VideoWriter(output_path, fourcc, 30.0, (frame_width, frame_height))
+
+    while True:
+        ret, frame = video_capture.read()
+        if not ret:
+            print("Видео закончилось.")
+            break
+
+        frame = cv.resize(frame, (frame_width, frame_height))
+        processed_frame = process_frame(frame, detector, recognizer, reference_features)
+        output_video.write(processed_frame)
+
+    video_capture.release()
+    output_video.release()
+    cv.destroyAllWindows()
+
+
+"""Основной процесс обработки видео."""
+detector, recognizer = load_models()
+reference_features = get_ground_base_features(detector, recognizer, GROUND_BASE_IMAGES)
+process_video(VIDEO_INPUT, VIDEO_OUTPUT, detector, recognizer, reference_features)
+print(f"Обработка завершена. Видео сохранено в {VIDEO_OUTPUT}")