Spaces:
Sleeping
Sleeping
File size: 3,235 Bytes
2c0da97 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 | import os
import pickle
import numpy as np
from deepface import DeepFace
import gc
# Configure models to match our Consensus Engine
MODELS = ["ArcFace", "Facenet512"]
DB_PATH = "faces_db"
PICKLE_PATH = "core/embeddings.pickle"
def generate_hard_pickle():
"""
Advanced Identity Encoder:
Creates a 'Centroid Vector' for each person. This is far more robust
to aging and lighting than single-image matching.
"""
print("🚀 Starting Advanced Face Encoding...")
if not os.path.exists(DB_PATH):
print(f"❌ Error: {DB_PATH} folder not found!")
return
# Structure: { "Name": { "ArcFace": [vec], "Facenet512": [vec] } }
identity_map = {}
# 1. Scan the database
image_files = [f for f in os.listdir(DB_PATH) if f.lower().endswith(('.jpg', '.jpeg', '.png'))]
if not image_files:
print("⚠️ No images found in faces_db!")
return
for filename in image_files:
# Extract name (e.g., 'Shashank_Kumar.jpg' -> 'Shashank_Kumar')
name = os.path.splitext(filename)[0].split('_')[0] # Groups by first name
img_path = os.path.join(DB_PATH, filename)
print(f"📸 Encoding: {name} ({filename})...")
if name not in identity_map:
identity_map[name] = {m: [] for n in MODELS}
try:
for model in MODELS:
# Extract 128/512 dimension embedding
# We use enforce_detection=True here to ensure we only pickle REAL faces
embedding_objs = DeepFace.represent(img_path=img_path, model_name=model,
enforce_detection=True, detector_backend="retinaface")
if embedding_objs:
vector = np.array(embedding_objs[0]["embedding"])
# L2 Normalization makes the vector age-resistant
norm_vector = vector / np.linalg.norm(vector)
identity_map[name][model].append(norm_vector)
except Exception as e:
print(f"⚠️ Skip {filename}: {e}")
# 2. Create the 'Master Identity Vectors' (Centroids)
final_embeddings = {}
for name, model_data in identity_map.items():
final_embeddings[name] = {}
for model, vectors in model_data.items():
if vectors:
# Average all vectors for this person to create a 'Super-Vector'
# This mathematically finds the 'Center' of their identity
centroid = np.mean(vectors, axis=0)
# Re-normalize the average
final_embeddings[name][model] = centroid / np.linalg.norm(centroid)
print(f"✅ Created Master Identity for {name} [{model}] using {len(vectors)} samples.")
# 3. Save the Hard Pickle
with open(PICKLE_PATH, "wb") as f:
pickle.dump(final_embeddings, f)
print(f"
✨ SUCCESS: Hard Pickle saved to {PICKLE_PATH}")
print("This file contains the mathematical essence of your students.")
gc.collect()
if __name__ == "__main__":
generate_hard_pickle()
|