Facial-feature-detector / parameters.yml
paresh95
PS | Add face comparison feature
988dde3
raw
history blame contribute delete
No virus
2.28 kB
face_detection:
config: "models/face_detection/deploy.prototxt.txt"
model: "models/face_detection/res10_300x300_ssd_iter_140000.caffemodel"
face_age:
config: "models/face_age/age_deploy.prototxt"
model: "models/face_age/age_net.caffemodel"
face_gender:
config: "models/face_gender/gender_deploy.prototxt"
model: "models/face_gender/gender_net.caffemodel"
face_landmarks:
model: 'models/face_alignment/shape_predictor_68_face_landmarks.dat'
results_interpretation:
Demographic predictions:
age_range: "Predicted age"
age_confidence: "Confidence of age prediction (0-1)"
gender: "Predicted gender"
gender_confidence: "Confidence of gender prediction (0-1)"
Face proportions:
top_of_nose_to_middle_of_mouth_vs_middle_mouth_to_bottom_of_chin: "Ideal value is 1.62 (golden ratio - from classical Greek work on maths and geometry)"
middle_of_mouth_to_bottom_of_mouth_vs_top_of_mouth_to_middle_of_mouth: "Ideal value is 1.62 (golden ratio - from classical Greek work on maths and geometry)"
eye_width_vs_distance_between_eyes: "Ideal value is 1"
eye_to_eyebrows_vs_eye_height: "Ideal value is 1"
center_of_left_to_right_eye_vs_mouth_width: "Ideal value is 1"
Face symmetry metrics:
structural_similarity: "Range: -1 (opposite) to 1 (similar). Considers differences in structural information, luminance, and texture."
cosine_distance: "Ranges: -1 to 1. 0 =similar, -1 or 1 = not similar. Considers differences in pixels."
mse: Lower means more similar. Minimum value is 0, maximum is unbounded. Measures pixel differences."
mae: Lower means more similar. Minimum value is 0, maximum is unbounded. Measures pixel differences."
histogram_correlation: "Range: -1 (opposite) to 1 (similar). Considers similarity based on color or intensity distributions."
histogram_intersection: "Range: 0 (similar) to total pixels in image (not similar). Consider similarity based on color or intensity distributions."
orb_detector_matches: "Higher is better. Counts the number of matches between keypoints in images."
pixel_difference: "Lower is better. Minimum value is 0, maximum is unbounded."
Face texture metrics:
texture_std: "Lower means less varied facial texture. Minimum is 0, maximum is unbounded."