added buffalo_l model to deepface

deepface/DeepFace.py

@@ -54,7 +54,7 @@ def build_model(model_name: str, task: str = "facial_recognition") -> Any:
     Args:
         model_name (str): model identifier
             - VGG-Face, Facenet, Facenet512, OpenFace, DeepFace, DeepID, Dlib,
-                ArcFace, SFace and GhostFaceNet for face recognition
+                ArcFace, SFace GhostFaceNet and buffalo_l for face recognition
             - Age, Gender, Emotion, Race for facial attributes
             - opencv, mtcnn, ssd, dlib, retinaface, mediapipe, yolov8, yolov11n,
               yolov11s, yolov11m, yunet, fastmtcnn or centerface for face detectors

deepface/models/facial_recognition/Buffalo_L.py

@@ -0,0 +1,73 @@
+import cv2
+import numpy as np
+from deepface.models.FacialRecognition import FacialRecognition
+from deepface.commons.logger import Logger
+from deepface.basemodel import get_weights_path
+from deepface.common import weight_utils
+import os
+
+logger = Logger()
+
+# Check for insightface dependency
+try:
+    from insightface.model_zoo import get_model
+except ModuleNotFoundError:
+    raise ModuleNotFoundError(
+        "InsightFace is an optional dependency for the Buffalo_L model."
+        "You can install it with: pip install insightface>=0.7.3"
+    )
+
+class Buffalo_L(FacialRecognition):
+    def __init__(self):
+        self.model = None
+        self.input_shape = (112, 112)  # Buffalo_L recognition model expects 112x112
+        self.output_shape = 512  # Embedding size for Buffalo_L
+        self.load_model()
+
+    def load_model(self):
+        root = os.path.join(get_weights_path(), 'insightface')
+        model_name = 'buffalo_l/w600k_r50.onnx'
+        model_path = os.path.join(root, model_name)
+
+        if not os.path.exists(model_path):
+            url = 'https://drive.google.com/file/d/1N0GL-8ehw_bz2eZQWz2b0A5XBdXdxZhg/view?usp=sharing'
+            weight_utils.download_file(url, model_path)
+
+        self.model = get_model(model_name, root=root)
+        self.model.prepare(ctx_id=-1, input_size=self.input_shape)
+
+    def preprocess(self, img):
+        """
+        Preprocess the image to match InsightFace recognition model expectations.
+        Args:
+            img (numpy array): Image in shape (1, 112, 112, 3) or (112, 112, 3)
+        Returns:
+            numpy array: Preprocessed image
+        """
+        if len(img.shape) == 4:  # (1, 112, 112, 3)
+            img = img[0]  # Remove batch dimension
+        if img.max() <= 1.0:  # If normalized to [0, 1]
+            img = (img * 255).astype(np.uint8)
+        # Always convert RGB to BGR (InsightFace expects BGR, DeepFace provides RGB)
+        img = img[:, :, ::-1]
+        return img
+
+    def forward(self, img):
+        """
+        Extract face embedding from a pre-cropped face image.
+        Args:
+            img (numpy array): Preprocessed face image with shape (1, 112, 112, 3)
+        Returns:
+            numpy array: Face embedding
+        """
+        img = self.preprocess(img)
+        embedding = self.model.get_feat(img)
+
+        # Handle different embedding formats
+        if isinstance(embedding, np.ndarray):
+            if len(embedding.shape) > 1:
+                embedding = embedding.flatten()
+        elif isinstance(embedding, list):
+            embedding = np.array(embedding).flatten()
+
+        return embedding

deepface/modules/modeling.py

@@ -11,7 +11,8 @@
     SFace,
     Dlib,
     Facenet,
-    GhostFaceNet
+    GhostFaceNet,
+    Buffalo_L
 )
 from deepface.models.face_detection import (
     FastMtCnn,
@@ -59,7 +60,8 @@ def build_model(task: str, model_name: str) -> Any:
             "Dlib": Dlib.DlibClient,
             "ArcFace": ArcFace.ArcFaceClient,
             "SFace": SFace.SFaceClient,
-            "GhostFaceNet": GhostFaceNet.GhostFaceNetClient
+            "GhostFaceNet": GhostFaceNet.GhostFaceNetClient,
+            "Buffalo_L": Buffalo_L.Buffalo_L
         },
         "spoofing": {
             "Fasnet": FasNet.Fasnet,

deepface/modules/verification.py

@@ -422,6 +422,7 @@ def find_threshold(model_name: str, distance_metric: str) -> float:
         "DeepFace": {"cosine": 0.23, "euclidean": 64, "euclidean_l2": 0.64},
         "DeepID": {"cosine": 0.015, "euclidean": 45, "euclidean_l2": 0.17},
         "GhostFaceNet": {"cosine": 0.65, "euclidean": 35.71, "euclidean_l2": 1.10},
+        "Buffalo_L": {"cosine": 0.65, "euclidean": 11.13, "euclidean_l2": 1.1},
     }
 
     threshold = thresholds.get(model_name, base_threshold).get(distance_metric, 0.4)

requirements.txt

@@ -12,4 +12,4 @@ flask_cors>=4.0.1
 mtcnn>=0.1.0
 retina-face>=0.0.14
 fire>=0.4.0
-gunicorn>=20.1.0
+gunicorn>=20.1.0

requirements_additional.txt

@@ -3,4 +3,7 @@ mediapipe>=0.8.7.3
 dlib>=19.20.0
 ultralytics>=8.0.122
 facenet-pytorch>=2.5.3
-torch>=2.1.2
+torch>=2.1.2
+insightface>=0.7.3
+onnxruntime>=1.9.0
+tf-keras