We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.CV

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo

Computer Science > Computer Vision and Pattern Recognition

Title: Canonical Face Embeddings

Abstract: We present evidence that many common convolutional neural networks (CNNs) trained for face verification learn functions that are nearly equivalent under rotation. More specifically, we demonstrate that one face verification model's embeddings (i.e. last-layer activations) can be compared directly to another model's embeddings after only a rotation or linear transformation, with little performance penalty. This finding is demonstrated using IJB-C 1:1 verification across the combinations of ten modern off-the-shelf CNN-based face verification models which vary in training dataset, CNN architecture, method of angular loss calculation, or some combination of the 3. These networks achieve a mean true accept rate of 0.96 at a false accept rate of 0.01. When instead evaluating embeddings generated from two CNNs, where one CNN's embeddings are mapped with a linear transformation, the mean true accept rate drops to 0.95 using the same verification paradigm. Restricting these linear maps to only perform rotation produces a mean true accept rate of 0.91. These mappings' existence suggests that a common representation is learned by models despite variation in training or structure. We discuss the broad implications a result like this has, including an example regarding face template security.
Comments: 13 pages, 9 figures, 3 tables
Subjects: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
Cite as: arXiv:2106.07822 [cs.CV]
  (or arXiv:2106.07822v3 [cs.CV] for this version)

Submission history

From: David McNeely-White [view email]
[v1] Tue, 15 Jun 2021 00:52:05 GMT (8979kb,D)
[v2] Thu, 17 Jun 2021 20:24:30 GMT (1846kb,D)
[v3] Wed, 27 Oct 2021 22:46:14 GMT (4079kb,D)

Link back to: arXiv, form interface, contact.