Merge pull request #1 from ShivamSinghal1/feature/facenet512-onnx

feat: added facenet512 onnx model support

deepface/models/facial_recognition/Facenet.py

@@ -1,5 +1,7 @@
 import os
 import gdown
+import numpy as np
+from typing import List
 from deepface.commons import package_utils, folder_utils
 from deepface.models.FacialRecognition import FacialRecognition
 from deepface.commons.logger import Logger
@@ -67,6 +69,24 @@ class FaceNet512dClient(FacialRecognition):
         self.output_shape = 512
 
 
+class FaceNet512dONNXClient(FacialRecognition):
+    """
+    FaceNet-1512d model class
+    """
+
+    def __init__(self):
+        self.model = load_facenet512d_onnx_model()
+        self.model_name = "FaceNet-512d-onnx"
+        self.input_shape = (160, 160)
+        self.output_shape = 512
+
+    def forward(self, img: np.ndarray) -> List[float]:
+        input_name = self.model.get_inputs()[0].name
+        output_name = self.model.get_outputs()[0].name
+        result = self.model.run([output_name], {input_name: img})
+        return result[0][0]
+
+
 def scaling(x, scale):
     return x * scale
 
@@ -1711,3 +1731,36 @@ def load_facenet512d_model(
     # -------------------------
 
     return model
+
+
+def load_facenet512d_onnx_model(
+    url="https://github.com/ShivamSinghal1/deepface/releases/download/v1/facenet512_fp32.onnx",
+) -> Model:
+    """
+    Download Facenet512d ONNX model weights and load
+    Returns:
+        model (Model)
+    """
+    try:
+        import onnxruntime as ort
+    except ModuleNotFoundError as e:
+        raise ImportError(
+            "FaceNet512dONNX is an optional model, ensure the library is installed. "
+            "Please install using 'pip install onnxruntime' or 'pip install onnxruntime-gpu' to use gpu"
+        ) from e
+
+    if ort.get_device() == "GPU":
+        logger.info(f"using onnx GPU for inference")
+        providers = ['CUDAExecutionProvider', 'CPUExecutionProvider']
+    else:
+        providers = ['CPUExecutionProvider']
+
+    home = folder_utils.get_deepface_home()
+    onnx_model_path = os.path.join(home, ".deepface/weights/facenet512_onnx_weights.onnx")
+
+    if not os.path.isfile(onnx_model_path):
+        logger.info(f"{os.path.basename(onnx_model_path)} will be downloaded...")
+        gdown.download(url, onnx_model_path, quiet=False)
+
+    model = ort.InferenceSession(onnx_model_path, providers=providers)
+    return model

deepface/modules/modeling.py

@@ -54,6 +54,7 @@ def build_model(task: str, model_name: str) -> Any:
             "OpenFace": OpenFace.OpenFaceClient,
             "Facenet": Facenet.FaceNet128dClient,
             "Facenet512": Facenet.FaceNet512dClient,
+            "Facenet512ONNX": Facenet.FaceNet512dONNXClient,
             "DeepFace": FbDeepFace.DeepFaceClient,
             "DeepID": DeepID.DeepIdClient,
             "Dlib": Dlib.DlibClient,

requirements_additional.txt

@@ -4,3 +4,4 @@ dlib>=19.20.0
 ultralytics>=8.0.122
 facenet-pytorch>=2.5.3
 torch>=2.1.2
+onnxruntime>=1.19.0

tests/test_verify.py

@@ -8,7 +8,7 @@ from deepface.commons.logger import Logger
 
 logger = Logger()
 
-models = ["VGG-Face", "Facenet", "Facenet512", "ArcFace", "GhostFaceNet"]
+models = ["VGG-Face", "Facenet", "Facenet512", "ArcFace", "GhostFaceNet", "Facenet512ONNX"]
 metrics = ["cosine", "euclidean", "euclidean_l2"]
 detectors = ["opencv", "mtcnn"]