How far can you get with a modern face recognition test set using only simple features?

By Nicolas Pinto, James J. DiCarlo, David D. Cox. 2009

244 citations recorded by the source corpus.

In recent years, large databases of natural images have become increasingly popular in the evaluation of face and object recognition algorithms. However, Pinto et al. previously illustrated an inherent danger in using such sets, showing that an extremely basic recognition system, built on a trivial feature set, was able to take advantage of low-level regularities in popular object and face recognition sets, performing on par with many state-of-the-art systems. Recently, several groups have raised the performance “bar” for these sets, using more advanced classification tools. However, it is difficult to know whether these improvements are due to progress towards solving the core computational problem, or are due to further improvements in the exploitation of low-level regularities. Here, we show that even modest optimization of the simple model introduced by Pinto et al. using modern multiple kernel learning (MKL) techniques once again yields “state-of-the-art” performance levels on a standard face recognition set (“labeled faces in the wild”). However, at the same time, even with the inclusion of MKL techniques, systems based on these simple features still fail on a synthetic face recognition test that includes more “realistic” view variation by design. These results underscore the importance of building test sets focussed on capturing the central computational challenges of real-world face recognition.

Subfields: Computational Mechanics, Computer Vision and Pattern Recognition

How researchers see this paper

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

In Computational Mechanics

In Computer Vision and Pattern Recognition

Citations in the corpus

Strike (With) a Pose: Neural Networks Are Easily Fooled by Strange Poses of Familiar Objects (2018, methodology): “Previous research has successfully used 3D graphics as a diagnostic tool for computer vision systems [7, 31, 47, 32, 50].”
How does the brain solve visual object recognition? (2012, neutral): “However, we and our collaborators recently used rapidly advancing computing power to build many thousands of algorithms, in which a very large set of operating parameters was learned (unsupervised) from naturalistic video (Pinto et al., 2009b).”
A High-Throughput Screening Approach to Discovering Good Forms of Biologically Inspired Visual Representation (2009, neutral): “rendered boats and animals 3) a set of rendered images of two synthetic faces (one male, one female, [10,12]), and 4) a modified”
Similarity Scores Based on Background Samples (2009, methodology): “These methods include the combined nowak+Merl system [29], the Nowak method [27], the hybrid method of [10], and the recent V1-like/mkl method of [33].”
Space-Variant Descriptor Sampling for Action Recognition Based on Saliency and Eye Movements (2012, unknown): “[17]).”
Face Recognition Across Non-Uniform Motion Blur, Illumination, and Pose (2015, background): “However, as pointed out in [38] [39], the”
Fisher Vector Faces in the Wild (2013, unknown): “V1-like/MKL [26] 0.”
Probabilistic Elastic Matching for Pose Variant Face Verification (2013, methodology): “[21], which achieved an average accuracy of 0.”
Describable Visual Attributes for Face Verification and Image Search (2011, unknown): “10 shows results on LFW for our attribute classifiers (red line), simile classifiers (blue line), and a hybrid of the two (green line), along with several previous methods (dotted lines) [27], [48], [56], [57], [60], [61].”
Tom-vs-Pete Classifiers and Identity-Preserving Alignment for Face Verification (2012, methodology): “Funneled images provided by the LFW maintainers have been widely used [12, 17, 26, 32, 35], but the Tom-vs-Pete classifiers require a correspondence more precise than funneling provides.”
Learning hierarchical representations for face verification with convolutional deep belief networks (2012, unknown): “Method μ̂± SE V1-like with MKL [25] 0.”
Deep Learning for Face Veriﬁcation (2013, methodology): “This is notably better than the eigen-faces benchmark, and on par with an algorithm using specialized kernels on raw pixels [7].”
Effective Unconstrained Face Recognition by Combining Multiple Descriptors and Learned Background Statistics (2011, unknown): “Since its recent publication, a lot of attention has been focused on improving performance on the benchmarks associated with the LFW database (see, e.g., [1], [2], [3], [6], [14], [15], [16], [17]).”
Partial Face Recognition: Alignment-Free Approach (2013, methodology): “The proposed MKD-SRC-GTP method outperforms the commercial matcher FaceVACS and the best published method (V1-like/MKL) [71], which strictly follows the LFW image-restricted training protocol.”
Face recognition with learning-based descriptor (2010, background): “LDML, funneled [9] Hybrid, aligned [20] V1−like, funneled[19] LBP, baseline Attribute [13] Simile [13] Attribute + Simile [13] Background sample [31] Multiple LE + comp (our best)”