diff --git a/deepface/modules/detection.py b/deepface/modules/detection.py index c31a026..f413f60 100644 --- a/deepface/modules/detection.py +++ b/deepface/modules/detection.py @@ -90,6 +90,22 @@ def extract_faces( height, width, _ = img.shape + def is_valid_landmark(coord: Optional[Union[tuple, list]]) -> bool: + """ + Check if a landmark coordinate is within valid image bounds. + + Args: + coord (tuple or list or None): (x, y) coordinate to check. + Returns: + bool: True if coordinate is valid and within bounds, False otherwise. + """ + if coord is None: + return False + if not (isinstance(coord, (tuple, list)) and len(coord) == 2): + return False + x, y = coord + return 0 <= x < width and 0 <= y < height + base_region = FacialAreaRegion(x=0, y=0, w=width, h=height, confidence=0) if detector_backend == "skip": @@ -149,22 +165,36 @@ def extract_faces( w = min(width - x - 1, int(current_region.w)) h = min(height - y - 1, int(current_region.h)) + landmarks = { + "left_eye":current_region.left_eye, + "right_eye":current_region.right_eye, + "nose":current_region.nose, + "mouth_left":current_region.mouth_left, + "mouth_right":current_region.mouth_right + } + + # Sanitize landmarks - set invalid ones to None + for key, value in landmarks.items(): + if not is_valid_landmark(value): + landmarks[key] = None + + facial_area = { "x": x, "y": y, "w": w, "h": h, - "left_eye": current_region.left_eye, - "right_eye": current_region.right_eye, + "left_eye": landmarks["left_eye"], + "right_eye": landmarks["right_eye"], } # optional nose, mouth_left and mouth_right fields are coming just for retinaface if current_region.nose is not None: - facial_area["nose"] = current_region.nose + facial_area["nose"] = landmarks["nose"] if current_region.mouth_left is not None: - facial_area["mouth_left"] = current_region.mouth_left + facial_area["mouth_left"] = landmarks["mouth_left"] if current_region.mouth_right is not None: - facial_area["mouth_right"] = current_region.mouth_right + facial_area["mouth_right"] = landmarks["mouth_right"] resp_obj = { "face": current_img, diff --git a/tests/test_landmark_sanitization.py b/tests/test_landmark_sanitization.py new file mode 100644 index 0000000..1cdf483 --- /dev/null +++ b/tests/test_landmark_sanitization.py @@ -0,0 +1,70 @@ +import numpy as np +import pytest +from deepface.modules.detection import extract_faces, DetectedFace, FacialAreaRegion +from deepface.commons.logger import Logger + +logger = Logger() + +def is_valid_landmark(coord, width, height): + if coord is None: + return False + if not (isinstance(coord, (tuple, list)) and len(coord) == 2): + return False + x, y = coord + return 0 <= x < width and 0 <= y < height + +def sanitize_landmarks(region, width, height): + landmarks = { + "left_eye": region.left_eye, + "right_eye": region.right_eye, + "nose": region.nose, + "mouth_left": region.mouth_left, + "mouth_right": region.mouth_right, + } + for key, value in landmarks.items(): + if not is_valid_landmark(value, width, height): + landmarks[key] = None + return landmarks + +def test_sanitize_landmarks(): + img = np.zeros((100, 100, 3), dtype=np.uint8) + height, width = img.shape[:2] + region = FacialAreaRegion( + x=10, y=10, w=50, h=50, + left_eye=(-5, 20), # invalid + right_eye=(20, 200), # invalid + nose=(30, 30), # valid + mouth_left=(150, 20), # invalid + mouth_right=(20, -10), # invalid + confidence=0.9 + ) + landmarks = sanitize_landmarks(region, width, height) + logger.info(f"Sanitized landmarks: {landmarks}") + assert landmarks["left_eye"] is None + assert landmarks["right_eye"] is None + assert landmarks["nose"] == (30, 30) + assert landmarks["mouth_left"] is None + assert landmarks["mouth_right"] is None + logger.info("Test passed: Invalid landmarks are sanitized to None.") + +def test_extract_faces_sanitizes_landmarks(monkeypatch): + img = np.zeros((100, 100, 3), dtype=np.uint8) + facial_area = FacialAreaRegion( + x=10, y=10, w=50, h=50, + left_eye=(-5, 20), # invalid + right_eye=(20, 200), # invalid + nose=(30, 30), # valid + mouth_left=(150, 20), # invalid + mouth_right=(20, -10), # invalid + confidence=0.9 + ) + detected_face = DetectedFace(img=img, facial_area=facial_area, confidence=0.9) + monkeypatch.setattr("deepface.modules.detection.detect_faces", lambda *args, **kwargs: [detected_face]) + result = extract_faces(img, detector_backend="opencv", enforce_detection=False) + facial_area_out = result[0]["facial_area"] + logger.info(f"Output facial_area: {facial_area_out}") + assert facial_area_out["left_eye"] is None + assert facial_area_out["right_eye"] is None + assert facial_area_out.get("nose") == (30, 30) + assert facial_area_out.get("mouth_left") is None + assert facial_area_out.get("mouth_right") is None \ No newline at end of file