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: Multi-Head Self-Attention via Vision Transformer for Zero-Shot Learning

Abstract: Zero-Shot Learning (ZSL) aims to recognise unseen object classes, which are not observed during the training phase. The existing body of works on ZSL mostly relies on pretrained visual features and lacks the explicit attribute localisation mechanism on images. In this work, we propose an attention-based model in the problem settings of ZSL to learn attributes useful for unseen class recognition. Our method uses an attention mechanism adapted from Vision Transformer to capture and learn discriminative attributes by splitting images into small patches. We conduct experiments on three popular ZSL benchmarks (i.e., AWA2, CUB and SUN) and set new state-of-the-art harmonic mean results {on all the three datasets}, which illustrate the effectiveness of our proposed method.
Comments: Irish Machine Vision and Image Processing Conference (IMVIP) 2021
Subjects: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
Cite as: arXiv:2108.00045 [cs.CV]
  (or arXiv:2108.00045v1 [cs.CV] for this version)

Submission history

From: Anjan Dutta [view email]
[v1] Fri, 30 Jul 2021 19:08:44 GMT (8034kb,D)

Link back to: arXiv, form interface, contact.