Wing Loss for Robust Facial Landmark Localisation with Convolutional Neural Networks

By Zhenhua Feng, Josef Kittler, Muhammad Awais, Patrik Huber, Xiaojun Wu. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2017

464 citations recorded by the source corpus.

We present a new loss function, namely Wing loss, for robust facial landmark localisation with Convolutional Neural Networks (CNNs). We first compare and analyse different loss functions including L2, L1 and smooth L1. The analysis of these loss functions suggests that, for the training of a CNN-based localisation model, more attention should be paid to small and medium range errors. To this end, we design a piece-wise loss function. The new loss amplifies the impact of errors from the interval (-w, w) by switching from L1 loss to a modified logarithm function. To address the problem of under-representation of samples with large out-of-plane head rotations in the training set, we propose a simple but effective boosting strategy, referred to as pose-based data balancing. In particular, we deal with the data imbalance problem by duplicating the minority training samples and perturbing them by injecting random image rotation, bounding box translation and other data augmentation approaches. Last, the proposed approach is extended to create a two-stage framework for robust facial landmark localisation. The experimental results obtained on AFLW and 300W demonstrate the merits of the Wing loss function, and prove the superiority of the proposed method over the state-of-the-art approaches.

Subfields: Computer Vision and Pattern Recognition, Signal Processing

View Paper: https://www.semanticscholar.org/paper/822346af4fafcc54d7d95da2cbbb0bbe189405b4

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

High-Resolution Representations for Labeling Pixels and Regions (2019, methodology): “The dataset [85] is a combination of HELEN [53], [28] ResNet-50 1 .”
Adaptive Wing Loss for Robust Face Alignment via Heatmap Regression (2019, methodology): “[18] is able to improve the above loss functions by proposing Wing loss that has constant gradient when error is large, and large gradient when the error is small.”
Face identity disentanglement via latent space mapping (2020, methodology): “Elnd is implemented using a pre-trained landmarks regression network [Feng et al. 2018], trained to regress 68 facial keypoints.”
Deep High-Resolution Representation Learning for Visual Recognition (2019, methodology): “Our approach performs slightly worse than LAB that uses extra boundary information [149] and PDB [40] that uses stronger data augmentation.”
RetinaFace: Single-stage Dense Face Localisation in the Wild (2019, methodology): “Nevertheless, in this paper we refer to a broader deﬁnition of face localisation which includes face detection [39], face alignment [13], pixel-wise face parsing [48] and 3D dense correspondence regression [2, 12].”
Exclusivity-Consistency Regularized Knowledge Distillation for Face Recognition (2020, methodology): “We detect the faces by adopting the FaceBoxes detector [58, 57] and localize ﬁve landmarks (two eyes, nose tip and two mouth corners) through a simple 6-layer CNN [11].”
Disentangling in Latent Space by Harnessing a Pretrained Generator (2020, methodology): “Elnd is implemented using a pre-trained landmarks regression network [17], trained to regress 68 facial keypoints.”
3FabRec: Fast Few-Shot Face Alignment by Reconstruction (2019, methodology): “Since inference complexity is equivalent to two forward-passes through a ResNet-18, our method is able to run at frame rates of close to 300fps on a TitanX GPU - an order of magnitude faster than state-of-the-art approaches with similar, high accuracy (LAB [53]=16fps, Wing [18]=30fps, Deep Regressio”
CNN-GCN Aggregation Enabled Boundary Regression for Biomedical Image Segmentation (2020, methodology): “Inspired by Wing-loss [5] and Smooth-L1 loss [10], we propose a new loss function (ﬁgures are shown in supplementary material) that can prevent the model from taking large update steps when reaching small range errors in the late training stage and can recover quickly when dealing with large errors…”
General Facial Representation Learning in a Visual-Linguistic Manner (2021, methodology): “Instead of using those complex loss functions designed by [30, 42, 97], we simply train the head with a soft-label cross-entropy loss.”