Advanced search
2 files | 17.62 MB Add to list

RADAM : texture recognition through randomized aggregated encoding of deep activation maps

Author
Organization
Project
Abstract
Texture analysis is a classical yet challenging task in computer vision for which deep neural networks are actively being applied. Most approaches are based on building feature aggregation modules around a pre-trained backbone and then fine-tuning the new architecture on specific texture recognition tasks. Here we propose a new method named Random encoding of Aggregated Deep Activation Maps (RADAM) which extracts rich texture representations without ever changing the backbone. The technique consists of encoding the output at different depths of a pre-trained deep convolutional network using a Random-ized Autoencoder (RAE). The RAE is trained locally to each image using a closed-form solution, and its decoder weights are used to compose a 1-dimensional texture representation that is fed into a linear SVM. This means that no fine-tuning or backpropagation is needed for the backbone. We explore RADAM on several texture benchmarks and achieve state-of-the-art results with different com putational budgets. Our results suggest that pre-trained backbones may not require additional fine-tuning for texture recog-nition if their learned representations are better encoded. & COPY; 2023 Elsevier Ltd. All rights reserved.
Keywords
Texture analysis, Randomized neural networks, Transfer learning, Convolutional networks, Feature extraction, CLASSIFICATION

Downloads

  • (...).pdf
    • full text (Published version)
    • |
    • UGent only
    • |
    • PDF
    • |
    • 2.72 MB
  • KERMIT-A1-722-accepted.pdf
    • full text (Accepted manuscript)
    • |
    • open access
    • |
    • PDF
    • |
    • 14.90 MB

Citation

Please use this url to cite or link to this publication:

MLA
Scabini, Leonardo, et al. “RADAM : Texture Recognition through Randomized Aggregated Encoding of Deep Activation Maps.” PATTERN RECOGNITION, vol. 143, 2023, doi:10.1016/j.patcog.2023.109802.
APA
Scabini, L., Zielinski, K. M., Ribas, L. C., Goncalves, W. N., De Baets, B., & Martinez Bruno, O. (2023). RADAM : texture recognition through randomized aggregated encoding of deep activation maps. PATTERN RECOGNITION, 143. https://doi.org/10.1016/j.patcog.2023.109802
Chicago author-date
Scabini, Leonardo, Kallil M. Zielinski, Lucas C. Ribas, Wesley N. Goncalves, Bernard De Baets, and Odemir Martinez Bruno. 2023. “RADAM : Texture Recognition through Randomized Aggregated Encoding of Deep Activation Maps.” PATTERN RECOGNITION 143. https://doi.org/10.1016/j.patcog.2023.109802.
Chicago author-date (all authors)
Scabini, Leonardo, Kallil M. Zielinski, Lucas C. Ribas, Wesley N. Goncalves, Bernard De Baets, and Odemir Martinez Bruno. 2023. “RADAM : Texture Recognition through Randomized Aggregated Encoding of Deep Activation Maps.” PATTERN RECOGNITION 143. doi:10.1016/j.patcog.2023.109802.
Vancouver
1.
Scabini L, Zielinski KM, Ribas LC, Goncalves WN, De Baets B, Martinez Bruno O. RADAM : texture recognition through randomized aggregated encoding of deep activation maps. PATTERN RECOGNITION. 2023;143.
IEEE
[1]
L. Scabini, K. M. Zielinski, L. C. Ribas, W. N. Goncalves, B. De Baets, and O. Martinez Bruno, “RADAM : texture recognition through randomized aggregated encoding of deep activation maps,” PATTERN RECOGNITION, vol. 143, 2023.
@article{01HGZ86MRAW431AVNSX4D1FW44,
  abstract     = {{Texture analysis is a classical yet challenging task in computer vision for which deep neural networks are actively being applied. Most approaches are based on building feature aggregation modules around a pre-trained backbone and then fine-tuning the new architecture on specific texture recognition tasks. Here we propose a new method named Random encoding of Aggregated Deep Activation Maps (RADAM) which extracts rich texture representations without ever changing the backbone. The technique consists of encoding the output at different depths of a pre-trained deep convolutional network using a Random-ized Autoencoder (RAE). The RAE is trained locally to each image using a closed-form solution, and its decoder weights are used to compose a 1-dimensional texture representation that is fed into a linear SVM. This means that no fine-tuning or backpropagation is needed for the backbone. We explore RADAM on several texture benchmarks and achieve state-of-the-art results with different com putational budgets. Our results suggest that pre-trained backbones may not require additional fine-tuning for texture recog-nition if their learned representations are better encoded. & COPY; 2023 Elsevier Ltd. All rights reserved.}},
  articleno    = {{109802}},
  author       = {{Scabini, Leonardo and Zielinski, Kallil M. and Ribas, Lucas C. and Goncalves, Wesley N. and De Baets, Bernard and Martinez Bruno, Odemir}},
  issn         = {{0031-3203}},
  journal      = {{PATTERN RECOGNITION}},
  keywords     = {{Texture analysis,Randomized neural networks,Transfer learning,Convolutional networks,Feature extraction,CLASSIFICATION}},
  language     = {{eng}},
  pages        = {{13}},
  title        = {{RADAM : texture recognition through randomized aggregated encoding of deep activation maps}},
  url          = {{http://doi.org/10.1016/j.patcog.2023.109802}},
  volume       = {{143}},
  year         = {{2023}},
}

Altmetric
View in Altmetric
Web of Science
Times cited: