Face Detection & Recognition

Module: Computer Vision | Difficulty: Intermediate

Face Detection Pipeline

MTCNN Architecture

Three-stage cascade:

1. P-Net: Proposal network for candidate windows
2. R-Net: Refine network to reject false positives
3. O-Net: Output network for facial landmarks

FaceNet Loss

where is anchor, is positive, is negative, is margin.

ArcFace Loss

Face Verification

import torch
import torch.nn as nn

class ArcFaceLoss(nn.Module):
    def __init__(self, embedding_dim, num_classes, s=30.0, m=0.50):
        super().__init__()
        self.s = s
        self.m = m
        self.weight = nn.Parameter(torch.randn(num_classes, embedding_dim))
        nn.init.xavier_uniform_(self.weight)
    
    def forward(self, embeddings, labels):
        cos_theta = nn.functional.linear(
            nn.functional.normalize(embeddings),
            nn.functional.normalize(self.weight)
        )
        theta = torch.acos(cos_theta.clamp(-1 + 1e-7, 1 - 1e-7))
        target_logits = torch.cos(theta + self.m)
        
        one_hot = torch.zeros_like(cos_theta)
        one_hot.scatter_(1, labels.view(-1, 1), 1.0)
        
        logits = one_hot * target_logits + (1 - one_hot) * cos_theta
        logits *= self.s
        return nn.functional.cross_entropy(logits, labels)

Key Takeaways

• MTCNN provides fast multi-scale face detection
• ArcFace angular margin improves discriminative power
• Anti-spoofing is critical for real-world deployment