File size: 4,571 Bytes
df53738 | 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 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 | """
CONSTABLE – Face detection and recognition engine.
Uses:
• MTCNN – fast face detection & alignment
• InceptionResnetV1 (pretrained='vggface2') – 512-d face embeddings
"""
import io
import base64
import logging
import numpy as np
from PIL import Image
logger = logging.getLogger(__name__)
# ─── Lazy imports so the app starts even if GPU is not available ───────────
try:
from facenet_pytorch import MTCNN, InceptionResnetV1
import torch
FACENET_OK = True
except ImportError:
FACENET_OK = False
logger.warning("facenet-pytorch not installed – face recognition disabled.")
try:
import cv2
CV2_OK = True
except ImportError:
CV2_OK = False
DEVICE = "cpu"
if FACENET_OK:
try:
import torch
if torch.cuda.is_available():
DEVICE = "cuda"
except Exception:
pass
_mtcnn = None
_resnet = None
def _get_models():
global _mtcnn, _resnet
if _mtcnn is None:
_mtcnn = MTCNN(
image_size=160,
margin=20,
min_face_size=40,
thresholds=[0.6, 0.7, 0.7],
factor=0.709,
post_process=True,
keep_all=False,
device=DEVICE,
)
if _resnet is None:
_resnet = InceptionResnetV1(pretrained="vggface2").eval().to(DEVICE)
return _mtcnn, _resnet
# ─── Public API ────────────────────────────────────────────────────────────
def decode_image(data_url: str) -> Image.Image:
"""Convert a base64 data-URL to a PIL Image (RGB)."""
if "," in data_url:
data_url = data_url.split(",", 1)[1]
raw = base64.b64decode(data_url)
img = Image.open(io.BytesIO(raw)).convert("RGB")
return img
def get_face_embedding(pil_image: Image.Image):
"""
Detect the largest face and return its 512-d embedding as a numpy array.
Returns (embedding: np.ndarray, face_crop: np.ndarray) or (None, None).
"""
if not FACENET_OK:
return None, None
mtcnn, resnet = _get_models()
try:
# MTCNN returns aligned face tensor (or None)
face_tensor, prob = mtcnn(pil_image, return_prob=True)
except Exception as e:
logger.debug(f"MTCNN error: {e}")
return None, None
if face_tensor is None:
return None, None
# Get the face crop as numpy for anti-spoofing
boxes, _ = mtcnn.detect(pil_image)
face_crop = None
if boxes is not None and len(boxes) > 0:
b = boxes[0].astype(int)
arr = np.array(pil_image)
x1, y1, x2, y2 = max(0, b[0]), max(0, b[1]), b[2], b[3]
face_crop = arr[y1:y2, x1:x2]
import torch
with torch.no_grad():
embedding = resnet(face_tensor.unsqueeze(0).to(DEVICE))
return embedding.squeeze().cpu().numpy(), face_crop
def get_embeddings_from_frames(data_urls: list):
"""
Process a list of base64 frame data-URLs.
Returns list of valid 512-d embeddings (may be empty).
"""
embeddings = []
for url in data_urls:
try:
img = decode_image(url)
emb, _ = get_face_embedding(img)
if emb is not None:
embeddings.append(emb.tolist())
except Exception as e:
logger.debug(f"Frame processing error: {e}")
return embeddings
def get_embeddings_and_crops_from_frames(data_urls: list):
"""
Process a list of base64 frame data-URLs.
Returns (embeddings: list of 512-d lists, face_crops: list of np.ndarray or None).
face_crops[i] is the face crop for frame i (None if no face in that frame).
"""
embeddings = []
crops = []
for url in data_urls:
try:
img = decode_image(url)
emb, face_crop = get_face_embedding(img)
if emb is not None:
embeddings.append(emb.tolist())
crops.append(face_crop)
else:
crops.append(None)
except Exception as e:
logger.debug(f"Frame processing error: {e}")
crops.append(None)
return embeddings, crops
def get_face_crops_from_frames(data_urls: list):
"""
Get face crops only from a list of base64 frame data-URLs (for liveness sequence).
Returns list of np.ndarray (face crops); frames with no face are omitted.
"""
_, crops = get_embeddings_and_crops_from_frames(data_urls)
return [c for c in crops if c is not None and c.size > 0]
|