75 lines
2.7 KiB
Plaintext
75 lines
2.7 KiB
Plaintext
#
|
|
# https://pypi.org/project/deepface/
|
|
#
|
|
|
|
from deepface import DeepFace
|
|
import os
|
|
import cv2
|
|
import numpy as np
|
|
|
|
# Function to compute embeddings for a given image path
|
|
def compute_embeddings(image_path, model_name='VGG-Face'):
|
|
try:
|
|
# This returns a list of embeddings for each face found in the image
|
|
embeddings = DeepFace.represent(img_path=image_path, model_name=model_name)
|
|
return [res['embedding'] for res in embeddings]
|
|
except Exception as e:
|
|
print(f"Error processing image {image_path}: {e}")
|
|
return []
|
|
|
|
# Function to find the best matches between reference and target images
|
|
def find_best_matches(reference_images, target_images, model_name='VGG-Face', metric='cosine'):
|
|
# Store embeddings for reference images
|
|
reference_embeddings = {}
|
|
|
|
for ref_img in reference_images:
|
|
ref_embeddings = compute_embeddings(ref_img, model_name=model_name)
|
|
if ref_embeddings:
|
|
reference_embeddings[ref_img] = ref_embeddings
|
|
|
|
# Store best matches
|
|
best_matches = []
|
|
|
|
# Loop through each target image
|
|
for target_img in target_images:
|
|
target_embeddings = compute_embeddings(target_img, model_name=model_name)
|
|
|
|
for target_emb in target_embeddings:
|
|
for ref_img, ref_emb_list in reference_embeddings.items():
|
|
# Compare each reference embedding with the target embedding
|
|
for ref_emb in ref_emb_list:
|
|
# Compute the distance between embeddings using DeepFace's distance functions
|
|
distance = DeepFace.find_distance(ref_emb, target_emb, distance_metric=metric)
|
|
|
|
# Store match details (target image, reference image, and distance)
|
|
best_matches.append({
|
|
'target_image': target_img,
|
|
'reference_image': ref_img,
|
|
'distance': distance
|
|
})
|
|
|
|
# Sort matches by the distance (smallest distances indicate best matches)
|
|
best_matches = sorted(best_matches, key=lambda x: x['distance'])
|
|
|
|
return best_matches
|
|
|
|
# Example usage
|
|
reference_images = [
|
|
"path_to_reference_image1.jpg",
|
|
"path_to_reference_image2.jpg"
|
|
]
|
|
|
|
target_images = [
|
|
"path_to_target_image1.jpg",
|
|
"path_to_target_image2.jpg",
|
|
# Add more target images here...
|
|
]
|
|
|
|
# Find the best matches
|
|
matches = find_best_matches(reference_images, target_images, model_name='VGG-Face', metric='cosine')
|
|
|
|
# Display the best match
|
|
for match in matches[:5]: # Display top 5 matches
|
|
print(f"Reference Image: {match['reference_image']}, Target Image: {match['target_image']}, Distance: {match['distance']}")
|
|
|