# Run inference embedding = session.run([output_name], input_name: img)[0]
# Convert to NCHW format (Batch, Channel, Height, Width) img = np.transpose(img, (2, 0, 1)) # HWC -> CHW img = np.expand_dims(img, axis=0) # Add batch dimension w600k-r50.onnx
# Normalize the embedding to unit length (cosine similarity) embedding = embedding / np.linalg.norm(embedding) # Run inference embedding = session
trtexec --onnx=w600k-r50.onnx --saveEngine=w600k-r50.engine --fp16 Intel's OpenVINO can quantize the model to FP16 for edge devices like the Intel NUC. """ # Convert BGR to RGB rgb = cv2
def get_face_embedding(face_image: np.ndarray) -> np.ndarray: """ face_image: BGR image from OpenCV, must be 112x112 pixels already cropped and aligned. Returns: 512-dim embedding vector. """ # Convert BGR to RGB rgb = cv2.cvtColor(face_image, cv2.COLOR_BGR2RGB)
Enter . At first glance, it looks like a cryptic filename. But to machine learning engineers and edge computing specialists, it represents a perfect balance of accuracy, speed, and portability.
return embedding.flatten() aligned_face = cv2.imread("aligned_face.jpg") embedding = get_face_embedding(aligned_face) print(f"Embedding shape: embedding.shape") # (512,) Step 3: Face Matching To verify if two faces belong to the same person: