Tom-vs-Pete Classifiers and Identity-Preserving Alignment for Face Verification

By Thomas Berg, Peter N. Belhumeur. 2012

168 citations recorded by the source corpus.

How researchers see this paper

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

1. Xiaogang Wang
2. Dacheng Tao
3. Xiaoou Tang
4. Ping Luo
5. Wanli Ouyang
6. Andrea Vedaldi
7. Andrew Zisserman
8. Yi Yang

Citations in the corpus

• Naive-Deep Face Recognition: Touching the Limit of LFW Benchmark or Not? (2015, background): “Extensive works have been done to push the accuracy limit on it [3, 16, 4, 1, 2, 5, 11, 10, 12, 14, 13, 17, 9].”
• Surpassing Human-Level Face Verification Performance on LFW with GaussianFace (2014, methodology): “2) Modern face verification methods are mainly divided into two categories: extracting low-level features [36, 3, 34, 10, 24], and building classification models [62, 50, 13, 37, 31, 56, 5, 28, 47, 33].”
• Deeply learned face representations are sparse, selective, and robust (2014, background): “Only very recently, deep learning achieved great success on face recognition [41, 31, 29, 28, 32] and significantly outperformed systems using low level features [18, 30, 11, 38, 14, 6, 1, 26, 7, 5].”
• Blessing of Dimensionality: High-Dimensional Feature and Its Efficient Compression for Face Verification (2013, background): “The second stage usually exploits supervised information to learn a classification model [10, 26, 30], discriminative subspace [3, 26, 36], or mid-level representation [4, 24, 34, 38].”
• Learning deep representations for ground-to-aerial geolocalization (2015, methodology): “Recent methods achieved high performances by extracting hand-crafted features at aligned fiducial points [5] [6] [8].”
• Multi-Directional Multi-Level Dual-Cross Patterns for Robust Face Recognition (2014, background): “, [26], [28], [29], [37], [38], MDML-DCPs also benefits from recent progress in face alignment algorithms [24], [25] that accurately locate dense facial feature points in real-time.”
• Face Recognition and Retrieval Using Cross-Age Reference Coding With Cross-Age Celebrity Dataset (2015, methodology): “[14] further improve the method by using “Tom-vs-Pete” classifier.”
• Fisher Vector Faces in the Wild (2013, background): “2) outperforms or performs just as well as the best face verification representations, including the ones that use elaborate face landmark detectors [3, 6] and multiple features [12].”
• Recover Canonical-View Faces in the Wild with Deep Neural Networks (2014, methodology): “Most of the best performing methods such as 凹， 11 , 7] belong to the second setting.”
• High-fidelity Pose and Expression Normalization for face recognition in the wild (2015, background): “However, most previous works either only keep the internal face region and dropping the information around the face [5, 23, 50, 29] or warp the pixels of surrounding region to the fixed positions so that some shape information is lost [9].”
• Learning Deep Face Representation (2014, background): “Pete+Attribute [18] LBP [2] Pyramid CNN (ours) Human on cropped faces”
• Sparsifying Neural Network Connections for Face Recognition (2015, methodology): “Our weight sparsifying algorithm is evaluated on the LFW dataset [17] and the YouTube Faces dataset [33], respectively, which has been extensively evaluated for face recognition in unconstrained conditions in recent years [5, 10, 23, 8, 30, 27, 26, 28, 31, 22].”
• A Comprehensive Survey on Pose-Invariant Face Recognition (2015, methodology): “…texture warping caused by pose variation, but they can save all the information in the original 2D image, for example, facial contour, hairstyle, clothes, and backgrounds [Berg and Belhumeur 2012; Taigman et al. 2014], which are also important cues for face recognition [Kumar et al. 2009].”
• Labeled Faces in the Wild: A Survey (2016, methodology): “5 Tom-vs-Pete Classifiers and Identity-Preserving Alignment for Face Verification, 2012 [16]”
• A Practical Transfer Learning Algorithm for Face Verification (2013, methodology): “We use the LFW dataset [13] for the target domain both because of its similarity with WDRef and because it represents a well-studied and challenging benchmark allowing us to place our results in the context of existing face verification systems [17, 26, 25, 2, 18, 6].”
• POOF: Part-Based One-vs.-One Features for Fine-Grained Categorization, Face Verification, and Attribute Estimation (2013, methodology): “The reference dataset, from [3], consists of 20,639 face images, downloaded from the internet, spanning 120 subject identities.”