An All-In-One Convolutional Neural Network for Face Analysis

By Rajeev Ranjan, S. Sankaranarayanan, Carlos Domingo Castillo, Ramalingam Chellappa. IEEE International Conference on Automatic Face & Gesture Recognition, 2016

450 citations recorded by the source corpus.

We present a multi-purpose algorithm for simultaneousface detection, face alignment, pose estimation, genderrecognition, smile detection, age estimation and face recognitionusing a single deep convolutional neural network (CNN). Theproposed method employs a multi-task learning framework thatregularizes the shared parameters of CNN and builds a synergyamong different domains and tasks. Extensive experimentsshow that the network has a better understanding of face andachieves state-of-the-art result for most of these tasks

Subfields: Computer Vision and Pattern Recognition, Signal Processing

View Paper: https://www.semanticscholar.org/paper/93420d9212dd15b3ef37f566e4d57e76bb2fab2f

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

Fine-Grained Head Pose Estimation Without Keypoints (2017, methodology): “All-InOne Convolutional Neural Network [21] for Face Analysis adds smile, age estimation and facial recognition to the former prediction tasks.”
Rank consistent ordinal regression for neural networks with application to age estimation (2019, background): “) can improve the overall performance across different tasks compared to a single-task CNN (Ranjan et al., 2017).”
Neural Photo Editing with Introspective Adversarial Networks (2016, background): “• Fiducial keypoint Error, measured as the mean L2 distance between the facial landmarks predicted by the system of (Sankaranarayanan et al., 2016).”
InclusiveFaceNet: Improving Face Attribute Detection with Race and Gender Diversity (2017, methodology): “…smile model SMILING + R + G reaches an accuracy 1.62% (absolute) higher than the 2016 ChaLearn Looking at People Challenge winner SIAT MMLAB (Zhang et al., 2016), and 0.13% (absolute) higher than the published state-of-the-art UMIACS (Rajeev Ranjan & Chellappa, 2017) on the FotW validation dataset.”
Transferability and Hardness of Supervised Classification Tasks (2019, background): “Training a single model to solve multiple tasks can be mutually beneficial to the individual tasks [23, 49, 62].”
Deep Facial Expression Recognition: A Survey (2018, methodology): “Furthermore, an all-in-one CNN model [138] was proposed to simultaneously solve a diverse set of face analysis tasks, including smile detection.”
MoFA: Model-Based Deep Convolutional Face Autoencoder for Unsupervised Monocular Reconstruction (2017, methodology): “In [40], a multi-task CNN is trained to predict several face-related parameters (e.”
Mitigating Bias in Gender, Age and Ethnicity Classification: A Multi-task Convolution Neural Network Approach (2018, background): “Automated FA is instrumental in a wide range of applications including face detection [2], soft biometrics / face attribute classification [3] and face recognition [4].”
Improving Smiling Detection with Race and Gender Diversity (2017, methodology): “UMIACS [28] fine-tuned their gender model on FotW train dataset before evaluating against FotW validation dataset.”
L2-constrained Softmax Loss for Discriminative Face Verification (2017, methodology): “On the other hand, the features obtained with L2-softmax loss have lower intra-class angular variability, and are represented by thinner lobes.”
Face recognition accuracy of forensic examiners, superrecognizers, and face recognition algorithms (2018, unknown): “Fusing individual examiners with A2017a and A2016 yielded performance equivalent to the fusion of two examiners (Mann–Whitney U test = 4.53× 104, n1 = 1,596, n2 =57, P=0.956; Mann–Whitney U test = 4.33× 104, n1 =1,596, n2 =57, P=0.526, respectively).”
UMDFaces: An annotated face dataset for training deep networks (2016, methodology): “The authors of [25] compare the performance of their keypoint detector with the performance of other algorithms and report state-of-the-art results on AFLW (Table II in [25]).”
Learning Pose-Aware Models for Pose-Invariant Face Recognition in the Wild (2019, background): “Finally, we note that other recent state-of-the-art IJB-A results reported after submission of this paper can be found in [15], [16], [32], [59], [60], [61], [62].”
A Modulation Module for Multi-task Learning with Applications in Image Retrieval (2018, background): “With the recent prevalence of deep learning-based approaches in various computer vision tasks, multitask learning is often implemented as parameter sharing in certain intermediate layers in a unified convolutional neural network architecture [4,5].”
Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification (2018, background): “…perception tasks including, but not limited to, face detection (Zafeiriou et al., 2015; Mathias et al., 2014; Bai and Ghanem, 2017), face classiﬁcation (Reid et al., 2013; Levi and Hassner, 2015a; Rothe et al., 2016) and face recognition (Parkhi et al., 2015; Wen et al., 2016; Ranjan et al., 2017).”
FSA-Net: Learning Fine-Grained Structure Aggregation for Head Pose Estimation From a Single Image (2019, methodology): “Several methods [31, 32] propose to perform various related facial analysis tasks simultaneously using CNNs.”
SfSNet: Learning Shape, Reflectance and Illuminance of Faces 'in the Wild' (2017, methodology): “For real images we detect keypoints using [23] and create a mask based on these keypoints.”
Self-Supervised Multi-level Face Model Learning for Monocular Reconstruction at Over 250 Hz (2017, unknown): “[47] proposed a multi-purpose CNN for regressing semantic parameters (e.”
Unsupervised Domain-Specific Deblurring via Disentangled Representations (2019, methodology): “The faces are detected and aligned using the method proposed in [32].”