Probabilistic Face Embeddings

Shi, Yong; Jain, Anil K.

doi:10.1109/iccv.2019.00700

Cited by 273 publications

(200 citation statements)

References 36 publications

Supporting

Mentioning

183

Contrasting

Order By: Relevance

“…The results in Table III show that RRAN can improve performance in most cases. In particular, at FAR=0.01%, RRAN outperforms PFE [49] by 0.94%. Based on the experimental results mentioned above, it can be seen that distance calibration scheme can improve the recognition performance robustly.…”

Section: Results On Youtube Faces Datasetmentioning

confidence: 95%

See 1 more Smart Citation

Redundancy Removing Aggregation Network With Distance Calibration for Video Face Recognition

Song

et al. 2021

IEEE Internet Things J.

View full text Add to dashboard Cite

Attention-based techniques have been successfully used for rating image quality, and have been widely employed for set-based face recognition. Nevertheless, for video face recognition, where the base Convolutional Neutral Network (CNN) trained on large-scale data already provides discriminative features, fusing features with only predicted quality scores to generate representation is likely to cause duplicate sample dominant problem, and degrade performance correspondingly. To resolve the problem mentioned above, we propose a Redundancy Removing Aggregation Network (RRAN) for video face recognition. Compared with other quality-aware aggregation schemes, RRAN can take advantage of similarity information to tackle the noise introduced by redundant video frames. By leveraging metric learning, RRAN introduces a distance calibration scheme to align distance distributions of negative pairs of different video representations, which improves the accuracy under a uniform threshold. A series of experiments are conducted on multiple realistic datasets to evaluate the performance of RRAN, including YouTube Faces, IJB-A, and IJB-C. In comprehensive experiments we demonstrate that our method can diminish the overall influence of poor quality components with large proportion in the video and further improve the overall recognition performance with individual difference. Specifically, RRAN achieves a 96.84% accuracy on YouTube Face, outperforming all existing aggregation schemes.

show abstract

Section: Results On Youtube Faces Datasetmentioning

confidence: 95%

“…[7] by 4.43% at 1% FPIR, and by 4.69% at 0.1% FAR, respectively. For IJB-C, we compute TAR at a lower FAR according to general practice [45]- [49]. The results in Table III show that RRAN can improve performance in most cases.…”

Section: Results On Youtube Faces Datasetmentioning

confidence: 99%

Redundancy Removing Aggregation Network With Distance Calibration for Video Face Recognition

Song

et al. 2021

IEEE Internet Things J.

View full text Add to dashboard Cite

show abstract

“…This work employs the provided model LResNet100E-IR. Probabilistic Face Embeddings (PFE) [ 45 ]. A recent state-of-the-art FR approach that represents the usual face embeddings as Gaussian distributions.…”

Section: Evaluation On Face Recognition Algorithmsmentioning

confidence: 99%

Improving the Head Pose Variation Problem in Face Recognition for Mobile Robots

Baltanas

Ruiz-Sarmiento

González-Jiménez

2021

Sensors

View full text Add to dashboard Cite

Face recognition is a technology with great potential in the field of robotics, due to its prominent role in human-robot interaction (HRI). This interaction is a keystone for the successful deployment of robots in areas requiring a customized assistance like education and healthcare, or assisting humans in everyday tasks. These unconstrained environments present additional difficulties for face recognition, extreme head pose variability being one of the most challenging. In this paper, we address this issue and make a fourfold contribution. First, it has been designed a tool for gathering an uniform distribution of head pose images from a person, which has been used to collect a new dataset of faces, both presented in this work. Then, the dataset has served as a testbed for analyzing the detrimental effects this problem has on a number of state-of-the-art methods, showing their decreased effectiveness outside a limited range of poses. Finally, we propose an optimization method to mitigate said negative effects by considering key pose samples in the recognition system’s set of known faces. The conducted experiments demonstrate that this optimized set of poses significantly improves the performance of a state-of-the-art, cutting-edge system based on Multitask Cascaded Convolutional Neural Networks (MTCNNs) and ArcFace.

show abstract

“…For any detector, the returned eye locations are used to perform normalization including a light face alignment and posterior cropping. This process involves a rotation and scaling, obtaining a final cropped image of 160 × 160 pixels where the eyes are located respectively in positions (52,52) and (104, 52).…”

Section: Development Of the Face Re-identification/verification Modulementioning

confidence: 99%

“…In our particular case, even when FaceNet is trained with our own dataset (https://github.com/ davidsandberg/facenet), we adopted the model (github.com/nyoki-mtl/keras-facenet) pretrained on the MS-Celeb-1M dataset [51], which has reported an accuracy of 99.6% for the Labeled Faces in the Wild dataset (LFW) [52]. Indeed, MS-Celeb-1M covers an extensive population of individuals, including one million faces from 100,000 users.…”

Section: Development Of the Face Re-identification/verification Modulementioning

confidence: 99%

Personal Guides: Heterogeneous Robots Sharing Personal Tours in Multi-Floor Environments

Rodriguez

Zabala

Marín-Reyes

et al. 2020

Sensors

View full text Add to dashboard Cite

GidaBot is an application designed to setup and run a heterogeneous team of robots to act as tour guides in multi-floor buildings. Although the tours can go through several floors, the robots can only service a single floor, and thus, a guiding task may require collaboration among several robots. The designed system makes use of a robust inter-robot communication strategy to share goals and paths during the guiding tasks. Such tours work as personal services carried out by one or more robots. In this paper, a face re-identification/verification module based on state-of-the-art techniques is developed, evaluated offline, and integrated into GidaBot’s real daily activities, to avoid new visitors interfering with those attended. It is a complex problem because, as users are casual visitors, no long-term information is stored, and consequently, faces are unknown in the training step. Initially, re-identification and verification are evaluated offline considering different face detectors and computing distances in a face embedding representation. To fulfil the goal online, several face detectors are fused in parallel to avoid face alignment bias produced by face detectors under certain circumstances, and the decision is made based on a minimum distance criterion. This fused approach outperforms any individual method and highly improves the real system’s reliability, as the tests carried out using real robots at the Faculty of Informatics in San Sebastian show.

show abstract

Probabilistic Face Embeddings

Cited by 273 publications

References 36 publications

Redundancy Removing Aggregation Network With Distance Calibration for Video Face Recognition

Redundancy Removing Aggregation Network With Distance Calibration for Video Face Recognition

Improving the Head Pose Variation Problem in Face Recognition for Mobile Robots

Personal Guides: Heterogeneous Robots Sharing Personal Tours in Multi-Floor Environments

Contact Info

Product

Resources

About