From Facial Parts Responses to Face Detection: A Deep Learning Approach

By Shuo Yang, Ping Luo, Chen Change Loy, Xiaoou Tang. 2015

590 citations recorded by the source corpus.

In this paper, we propose a novel deep convolutional network (DCN) that achieves outstanding performance on FDDB, PASCAL Face, and AFW. Specifically, our method achieves a high recall rate of 90.99% on the challenging FDDB benchmark, outperforming the state-of-the-art method [23] by a large margin of 2.91%. Importantly, we consider finding faces from a new perspective through scoring facial parts responses by their spatial structure and arrangement. The scoring mechanism is carefully formulated considering challenging cases where faces are only partially visible. This consideration allows our network to detect faces under severe occlusion and unconstrained pose variation, which are the main difficulty and bottleneck of most existing face detection approaches. We show that despite the use of DCN, our network can achieve practical runtime speed.

Subfields: Computer Vision and Pattern Recognition, Signal Processing

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

Unsupervised Cross-Domain Image Generation (2016, methodology): “…without this upscale are shown, for comparison, in Appendix B. Comparison With Human Annotators For evaluation purposes only, a team of professional an-notators manually created an emoji, using the web service of bitmoji.com , for 118 random images from the CelebA dataset (Yang et al., 2015).”
Residual Attention Network for Image Classification (2017, methodology): “Region proposal [26, 4, 8, 38] has been successfully adopted in image detection task.”
Emergence of Invariance and Disentanglement in Deep Representations (2017, methodology): “Table 1: (Left) The Small AlexNet model used in the random label experiment, adapted from Zhang et al. (2017). All convolutions have a 5×5 kernel.”
Fully-Adaptive Feature Sharing in Multi-Task Networks with Applications in Person Attribute Classification (2016, background): “It has also been shown that facial attribute prediction is helpful as an auxiliary task for improving face detection [47] and face alignment [50].”
PyramidBox: A Context-assisted Single Shot Face Detector (2018, methodology): “We evaluate our face detector on FDDB against the other state-of-the-art methods [20,4,39,24,40,18,41,42,43,44,45,46,47,48,49,50,51,52,29,53].”
Multimodal Generative Models for Scalable Weakly-Supervised Learning (2018, unknown): “Training and testing splits are replicated as in Yang et al. (2015).”
SSDMNV2: A real time DNN-based face mask detection system using single shot multibox detector and MobileNetV2 (2020, unknown): “These issues have been studied in some existing research papers such as (Ghiasi & Fowlkes, 2014; Opitz, Waltner, Poier, Possegger, & Bischof, 2016; Yang, Luo, Loy, & Tang, 2015) still, there is an excellent challenge for a vast dataset so that an efficient face mask detection model can be easily…”
Scale-Aware Face Detection (2017, background): “The CNN based approaches have recently shown great successes [13, 39, 4].”
Non-Adversarial Image Synthesis With Generative Latent Nearest Neighbors (2018, background): “To evaluate the performance of our method on higher resolution images, we trained our method on the CelebA-HQ dataset at 256 × 256 resolution.”
A survey of the recent architectures of deep convolutional neural networks (2019, result): “Zhang’s technique showed good results when compared to state-of-the-art techniques (Li et al. 2015; Ranjan et al. 2015; Yang et al. 2015).”
SSH: Single Stage Headless Face Detector (2017, methodology): “We compare SSH with HR [7], CMS-RCNN [38], Mul-titask Cascade CNN [37], LDCF [20], Faceness [34], and Multiscale Cascade CNN [35].”
Improving Person Re-identification by Attribute and Identity Learning (2017, methodology): “mprove the face alignment performance using the convolutional neural network. In [27], two CNN structures are cascaded and ﬁnetuned jointly with attribute tags to predict face attributes. Yang et al. [43] train CNNs for facial attribute recognition in order to obtain high responses in regions of fac”
Look at Boundary: A Boundary-Aware Face Alignment Algorithm (2018, background): “Different to face detection [45] and recognition [75], face alignment identifies geometry structure of human face which can be viewed as modeling highly structured output.”
WIDER FACE: A Face Detection Benchmark (2015, methodology): “The VJ [21], DPM [17], and Faceness [28] detectors are either obtained from the authors or from open source library (OpenCV).”
Finding Tiny Faces (2016, background): “Easy Medium Hard ACF[24] 0.659 0.541 0.273 Two-stage CNN[26] 0.681 0.618 0.323 Multiscale Cascade CNN[25] 0.691 0.634 0.345 Faceness[25] 0.713 0.664 0.424 Multitask Cascade CNN[27] 0.848 0.825 0.598 CMS-RCNN[28] 0 [7] and hard-example mining [21] to ameliorate such effects.”
One-Sided Unsupervised Domain Mapping (2017, background): “CelebA [26, 14] was annotated for multiple attributes including the person’s gender, hair color, and the existence of glasses in the image.”
Dynamic Multi-Task Learning with Convolutional Neural Network (2017, background): “…various fields, such as object detection [Girshick et al., 2014; Ren et al., 2015; Redmon et al., 2016], image classification [Krizhevsky et al., 2012; Szegedy et al., 2015; He et al., 2016] and face related problems [Sun et al., 2014; Taigman et al., 2014; Schroff et al., 2015; Yang et al., 2015].”
Multi-scale Fully Convolutional Network for Face Detection in the Wild (2017, methodology): “On the WIDER FACE dataset, we compare our MS-FCN detector against several state-of-the-art methods: CMS-RCNN [48], Multi-task Cascade CNN [45], Faceness-WIDER [40], Multi-Scale Cascade CNN [41], Two-Stage CNN [41], and ACF-WIDER [39].”