A New Dataset and Boundary-Attention Semantic Segmentation for Face Parsing

Liu, Yinglu; Shi, Hailin; Shen, Hao; Si, Yue; Wang, Xiaobo; Mei, Tao

doi:10.1609/aaai.v34i07.6832

Cited by 54 publications

(47 citation statements)

References 34 publications

(51 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The experiments are established on JD-landmark-mask dataset [9,10,11] for verification of the proposed method.…”

Section: Methodsmentioning

confidence: 99%

“…In the context of COVID-19 epidemic, all faces of this tournament are masked, which brings critical difficulties to the face alignment task. As augmentation, we use LaPa dataset [10] and open-source mask adding algorithm [11] to make more faces with various masks. Area-under-curve (AUC) from the cumulative errors distribution (CED) curve is used to evaluate the performance of Fig.…”

Section: Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Towards Efficient Masked-Face Alignment Via Cascaded Regression

Wen

Ding

Yao

et al. 2021

2021 IEEE International Conference on Multimedia &Amp; Expo Workshops (ICMEW)

View full text Add to dashboard Cite

Practical and efficient face alignment has been highly required and widely focused in recent years, especially under the trend of edge computation and real-time operation. And it is a critical need to deal with masked faces in the context of COVID-19 epidemic. In this paper, we propose a novel cascaded facial landmark detector towards efficient masked face alignment, which we call QCN (Quantized Cascaded Network). QCN consists of three stages: alignment, estimation and refinement. The alignment stage help to pre-align the faces to alleviate extreme poses. And the next two stages localize facial landmarks in a coarse-to-fine manner. Thanks to the Network Architecture Search and Quantization techniques, the networks of QCN are designed as efficient as possible. Specifically, QCN occupies 1.75 Mb storage and runs in 84.18 MFLOPs only. Despite costs little computations, the proposed method yields 62.62% AUC (@0.08) on test set of JD-landmark-mask, which achieves 2nd place in the Grand Challenge of 106-point Facial Landmark Localization in ICME2021.

show abstract

“…The experiments are established on JD-landmark-mask dataset [9,10,11] for verification of the proposed method.…”

Section: Methodsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Towards Efficient Masked-Face Alignment Via Cascaded Regression

Wen

Ding

Yao

et al. 2021

2021 IEEE International Conference on Multimedia &Amp; Expo Workshops (ICMEW)

View full text Add to dashboard Cite

show abstract

“…4 Images using different annotation protocols. Some of those protocols are listed from left to right, top to bottom: MUCT (76 points) [62], Helen (194 points) [63], LaPa (106 points) [64,65], AFLW (21 points) [66], Multi-PIE (68 points) [3] and MTFL (5 points) [67] in multiple situations. Partial occlusions caused by hair, glasses, face masks, self-touching, or some external occlusion objects, hide several face regions complicating the landmark evaluation.…”

Section: Robust and Fast Face Alignment Systemmentioning

confidence: 99%

Real-time face alignment: evaluation methods, training strategies and implementation optimization

Casado

López

2021

J Real-Time Image Proc

View full text Add to dashboard Cite

Face alignment is a crucial component in most face analysis systems. It focuses on identifying the location of several keypoints of the human faces in images or videos. Although several methods and models are available to developers in popular computer vision libraries, they still struggle with challenges such as insufficient illumination, extreme head poses, or occlusions, especially when they are constrained by the needs of real-time applications. Throughout this article, we propose a set of training strategies and implementations based on data augmentation, software optimization techniques that help in improving a large variety of models belonging to several real-time algorithms for face alignment. We propose an extended set of evaluation metrics that allow novel evaluations to mitigate the typical problems found in real-time tracking contexts. The experimental results show that the generated models using our proposed techniques are faster, smaller, more accurate, more robust in specific challenging conditions and smoother in tracking systems. In addition, the training strategy shows to be applicable across different types of devices and algorithms, making them versatile in both academic and industrial uses.

show abstract

“…We adopt a virtual mask adding algorithm in FaceX-Zoo [1] to put diverse masks in face images. Besides, we use a face parsing model trained on Lapa [4] dataset to recovery the facial occlusion. The face images are accompanied by a corresponding bounding box to ensure that only one is used if multiple faces are detected.…”

Section: Descriptionmentioning

confidence: 99%

The 3rd Grand Challenge of Lightweight 106-Point Facial Landmark Localization on Masked Faces

Xiang

Liu

Liao

et al. 2021

2021 IEEE International Conference on Multimedia &Amp; Expo Workshops (ICMEW)

Self Cite

View full text Add to dashboard Cite

Facial landmark localization is a fundamental component in various face related applications, such as face recognition, 3D face reconstruction, facial pose estimation, face synthesis and forensics, etc. However, due to the global pandemic of COVID-19, the facial mask has become a common way to suppress the transmission of the virus. This apparently makes conventional facial landmark localization unfaithful and inefficient because of occlusion. To lead the cutting-edge algorithms on masked faces, we host the 3rd grand challenge of lightweight 106-point facial landmark localization in conjunction with ICME 2021, aiming to improve the accuracy and robustness of facial landmark localization in real-world situations, especially on masked faces. Specifically, we construct a new dataset, named JD-landmark-mask, on the basis of the previous two competitions. It contains about 27,000 face images of two kinds, with real and virtual masks, which are largely varied in identity, head pose, facial expression, and occlusion. Besides, the strict limitations of model size (2M) and computational complexity (100M Flops) are set up for the lightweight model. Finally, more than 80 worldwide universities and research institutes took part in the competition. We will briefly introduce the information of the competition, as well as algorithms and results of the top three teams in this paper.

show abstract

A New Dataset and Boundary-Attention Semantic Segmentation for Face Parsing

Cited by 54 publications

References 34 publications

Towards Efficient Masked-Face Alignment Via Cascaded Regression

Towards Efficient Masked-Face Alignment Via Cascaded Regression

Real-time face alignment: evaluation methods, training strategies and implementation optimization

The 3rd Grand Challenge of Lightweight 106-Point Facial Landmark Localization on Masked Faces

Contact Info

Product

Resources

About