Dockerfile

@@ -26,8 +26,9 @@ RUN DEBIAN_FRONTEND=NONINTERACTIVE apt-get install -y \
 # Install the latest npm and npx
 RUN npm install --global npm@latest
 
-# Install deepface
+# Install deepface and retina-face
 RUN pip install deepface
+RUN pip install retina-face
 RUN pip install piexif
 
 # numpy 1.24 deprecated float; deepface is still using it, so we need to

ketrface/cluster.py

@@ -21,8 +21,6 @@ pictures_path = merge_config_path(config['path'], config['picturesPath'])
 faces_path = merge_config_path(config['path'], config['facesPath'])
 db_path = merge_config_path(config['path'], config["db"]["photos"]["host"])
 html_base = config['basePath']
-if html_base == "/":
-  html_base = "."
 
 # TODO
 # Switch to using DBSCAN

ketrface/detect.py

@@ -7,6 +7,8 @@ import argparse
 
 from PIL import Image, ImageOps
 from deepface import DeepFace
+from deepface.detectors import FaceDetector
+from retinaface import RetinaFace
 import numpy as np
 import cv2
 
@@ -22,8 +24,8 @@ faces_path = merge_config_path(config['path'], config['facesPath'])
 db_path = merge_config_path(config['path'], config["db"]["photos"]["host"])
 html_base = config['basePath']
 
-model_name = 'VGG-Face'
+model_name = 'VGG-Face'    # 'ArcFace'
-detector_backend = 'mtcnn'
+detector_backend = 'mtcnn' # 'retinaface'
 model = DeepFace.build_model(model_name)
 
 # Derived from
@@ -68,38 +70,44 @@ def variance_of_laplacian(image):
 def extract_faces(
   img, threshold=0.95, allow_upscaling = True, focus_threshold = 100):
 
-  img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) # mtcnn expects RGB
+  if detector_backend == 'retinaface':
+    faces = RetinaFace.detect_faces(
+      img_path = img, 
+      threshold = threshold, 
+      model = model, 
+      allow_upscaling = allow_upscaling)
+  elif detector_backend == 'mtcnn':
+    img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) # mtcnn expects RGB
 
-  redirect_on()
+    redirect_on()
-  res = face_detector.detect_faces(img_rgb)
+    res = face_detector.detect_faces(img_rgb)
-  redirect_off()
+    redirect_off()
 
-  if type(res) != list:
+    faces = {}
-    return None
+    if type(res) == list:
-
+      for i, face in enumerate(res):
-  faces = {}
+        if threshold > face['confidence']:
-  for i, face in enumerate(res):
+          continue
-    if threshold > face['confidence']:
+        x = face['box'][0]
-      continue
+        y = face['box'][1]
-    x = face['box'][0]
+        w = face['box'][2]
-    y = face['box'][1]
+        h = face['box'][3]
-    w = face['box'][2]
+        # If face is less than 2.5% of the image width and height,
-    h = face['box'][3]
+        # skip it (too small) -- filters out likely blurry faces in
-    # If face is less than 2.5% of the image width and height,
+        # large group photos where the actual face may exceed
-    # skip it (too small) -- filters out likely blurry faces in
+        # min_face_size passed to MTCNN
-    # large group photos where the actual face may exceed
+        if 0.025 > w / img.shape[0] and 0.025 > h / img.shape[1]:
-    # min_face_size passed to MTCNN
+          print(f'Dropping due to small face size: {w / img.shape[0]} x {h / img.shape[1]}')
-    if 0.025 > w / img.shape[0] and 0.025 > h / img.shape[1]:
+          continue
-      print(f'Dropping due to small face size: {w / img.shape[0]} x {h / img.shape[1]}')
+        faces[f'face_{i+1}'] = { # standardize properties
-      continue
+          'facial_area': [ x, y, x + w, y + h ],
-    faces[f'face_{i+1}'] = { # standardize properties
+          'landmarks': {
-      'facial_area': [ x, y, x + w, y + h ],
+            'left_eye': list(face['keypoints']['left_eye']),
-      'landmarks': {
+            'right_eye': list(face['keypoints']['right_eye']),
-        'left_eye': list(face['keypoints']['left_eye']),
+          },
-        'right_eye': list(face['keypoints']['right_eye']),
+          'score': face['confidence'],
-      },
+          
-      'score': face['confidence'],
+        }
-    }
 
   # Re-implementation of 'extract_faces' with the addition of keeping a 
   # copy of the face image for caching on disk
@@ -182,13 +190,8 @@ def extract_faces(
 
 
 parser = argparse.ArgumentParser(description = 'Detect faces in images.')
-parser.add_argument(
+parser.add_argument('photos', metavar='PHOTO', type=int, nargs='*',
-  'photos', 
+                    help='PHOTO ID to scan (default: all unscanned photos)')
-  metavar = 'PHOTO', 
-  type=int, 
-  nargs='*',
-  help = 'PHOTO ID to scan (default: all unscanned photos)'
-)
 args = parser.parse_args()
 print(args)
 
@@ -206,7 +209,7 @@ with conn:
   for i, row in enumerate(rows):
     photoId, photoFaces, albumPath, photoFilename = row
     img_path = f'{base}{albumPath}{photoFilename}'
-    print(f'Processing {i+1}/{count}: photoId = {photoId}: {img_path}')
+    print(f'Processing {i+1}/{count}: {img_path}')
     try:
       img = Image.open(img_path)
       img = ImageOps.exif_transpose(img) # auto-rotate if needed
@@ -244,8 +247,6 @@ with conn:
         'descriptorId': faceDescriptorId,
       })
 
-      print(f'Face added to database with faceId = {faceId}')
-
       path = f'{faces_path}/{"{:02d}".format(faceId % 100)}'
       try:
         os.makedirs(path)

ketrface/headers.py

@@ -7,18 +7,10 @@ import uu
 from io import BytesIO
 
 from ketrface.util import *
-from ketrface.config import *
-
-config = read_config()
-
-html_path = merge_config_path(config['path'], 'frontend')
-pictures_path = merge_config_path(config['path'], config['picturesPath'])
-faces_path = merge_config_path(config['path'], config['facesPath'])
-db_path = merge_config_path(config['path'], config["db"]["photos"]["host"])
-html_base = config['basePath']
 
+face_base = "../"
 faceId = int(sys.argv[1])
-path = f'{faces_path}/{"{:02d}".format(faceId % 100)}'
+path = f'{face_base}faces/{"{:02d}".format(faceId % 10)}'
 
 img = Image.open(f'{path}/{faceId}.jpg')
 exif_dict = piexif.load(img.info["exif"])