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 ScienceWISE logo

Computer Science > Computer Vision and Pattern Recognition

Title: Learnable Adaptive Cosine Estimator (LACE) for Image Classification

Abstract: In this work, we propose a new loss to improve feature discriminability and classification performance. Motivated by the adaptive cosine/coherence estimator (ACE), our proposed method incorporates angular information that is inherently learned by artificial neural networks. Our learnable ACE (LACE) transforms the data into a new "whitened" space that improves the inter-class separability and intra-class compactness. We compare our LACE to alternative state-of-the art softmax-based and feature regularization approaches. Our results show that the proposed method can serve as a viable alternative to cross entropy and angular softmax approaches. Our code is publicly available: this https URL
Comments: Accepted to WACV 2022; 14 pages (including appendix), 3 figures
Subjects: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Cite as: arXiv:2110.05324 [cs.CV]
  (or arXiv:2110.05324v3 [cs.CV] for this version)

Submission history

From: Joshua Peeples [view email]
[v1] Mon, 11 Oct 2021 14:45:15 GMT (6706kb,D)
[v2] Fri, 15 Oct 2021 14:17:35 GMT (6706kb,D)
[v3] Tue, 23 Nov 2021 16:00:07 GMT (6706kb,D)

Link back to: arXiv, form interface, contact.