SymFace: Additional Facial Symmetry Loss for Deep Face Recognition

TL;DR

SymFace introduces a symmetry-aware loss for deep face recognition that leverages vertical splits of frontal face images. The method defines a facial frontness score $\rho$ using landmark coordinates and uses a 3PSS algorithm to select symmetric images for split augmentation; a symmetry loss $L_\rho$ minimizes the embedding gap between left and right halves, with total loss $L_{total} = L_f + L_\rho$. This approach can be added on top of existing losses (e.g., ArcFace, AdaFace) and shows improvements across standard benchmarks (LFW, CFP-FP, CP-LFW, AgeDB, CA-LFW) on multiple backbones, indicating increased inter-class variance and more robust embeddings. The work demonstrates that incorporating natural facial symmetry can yield SoTA or near-SoTA results in frontal-view datasets and motivates further exploration for pose-variant data and efficient training.

Abstract

Over the past decade, there has been a steady advancement in enhancing face recognition algorithms leveraging advanced machine learning methods. The role of the loss function is pivotal in addressing face verification problems and playing a game-changing role. These loss functions have mainly explored variations among intra-class or inter-class separation. This research examines the natural phenomenon of facial symmetry in the face verification problem. The symmetry between the left and right hemi faces has been widely used in many research areas in recent decades. This paper adopts this simple approach judiciously by splitting the face image vertically into two halves. With the assumption that the natural phenomena of facial symmetry can enhance face verification methodology, we hypothesize that the two output embedding vectors of split faces must project close to each other in the output embedding space. Inspired by this concept, we penalize the network based on the disparity of embedding of the symmetrical pair of split faces. Symmetrical loss has the potential to minimize minor asymmetric features due to facial expression and lightning conditions, hence significantly increasing the inter-class variance among the classes and leading to more reliable face embedding. This loss function propels any network to outperform its baseline performance across all existing network architectures and configurations, enabling us to achieve SoTA results.

Figures (6)

• Figure 1: Symmetrical analysis on face dataset
• Figure 2: Images being evaluated by the 3PSS
• Figure 3: A comparison of $\rho$ various images from the dataset
• Figure 4: Face Image Splitting Process
• Figure 5: Workflow of split samples