Difference between revisions of "Publications/kirszenberg.21.dgmm"
From LRDE
(Created page with "{{Publication | published = true | date = 2021-02-16 | authors = Alexandre Kirszenberg, Guillaume Tochon, Élodie Puybareau, Jesus Angulo | title = Going beyond p-convolutions...") |
|||
Line 8: | Line 8: | ||
| address = Uppsala, Sweden |
| address = Uppsala, Sweden |
||
| publisher = Springer |
| publisher = Springer |
||
+ | | pages = 470 to 482 |
||
| abstract = Integrating mathematical morphology operations within deep neural networks has been subject to increasing attention lately. However, replacing standard convolution layers with erosions or dilations is particularly challenging because the min and max operations are not differentiable. Relying on the asymptotic behavior of the counter-harmonic mean, p-convolutional layers were proposed as a possible workaround to this issue since they can perform pseudo-dilation or pseudo-erosion operations (depending on the value of their inner parameter p), and very promising results were reported. In this work, we present two new morphological layers based on the same principle as the p-convolutional layer while circumventing its principal drawbacks, and demonstrate their potential interest in further implementations within deep convolutional neural network architectures. |
| abstract = Integrating mathematical morphology operations within deep neural networks has been subject to increasing attention lately. However, replacing standard convolution layers with erosions or dilations is particularly challenging because the min and max operations are not differentiable. Relying on the asymptotic behavior of the counter-harmonic mean, p-convolutional layers were proposed as a possible workaround to this issue since they can perform pseudo-dilation or pseudo-erosion operations (depending on the value of their inner parameter p), and very promising results were reported. In this work, we present two new morphological layers based on the same principle as the p-convolutional layer while circumventing its principal drawbacks, and demonstrate their potential interest in further implementations within deep convolutional neural network architectures. |
||
| lrdepaper = http://www.lrde.epita.fr/dload/papers/kirszie.2021.dgmm.pdf |
| lrdepaper = http://www.lrde.epita.fr/dload/papers/kirszie.2021.dgmm.pdf |
||
| lrdekeywords = Image |
| lrdekeywords = Image |
||
− | | note = Accepted |
||
| lrdenewsdate = 2021-02-16 |
| lrdenewsdate = 2021-02-16 |
||
| type = inproceedings |
| type = inproceedings |
||
| id = kirszenberg.21.dgmm |
| id = kirszenberg.21.dgmm |
||
+ | | identifier = doi:10.1007/978-3-030-76657-3_34 |
||
| bibtex = |
| bibtex = |
||
@InProceedings<nowiki>{</nowiki> kirszenberg.21.dgmm, |
@InProceedings<nowiki>{</nowiki> kirszenberg.21.dgmm, |
||
Line 28: | Line 29: | ||
address = <nowiki>{</nowiki>Uppsala, Sweden<nowiki>}</nowiki>, |
address = <nowiki>{</nowiki>Uppsala, Sweden<nowiki>}</nowiki>, |
||
publisher = <nowiki>{</nowiki>Springer<nowiki>}</nowiki>, |
publisher = <nowiki>{</nowiki>Springer<nowiki>}</nowiki>, |
||
+ | pages = <nowiki>{</nowiki>470--482<nowiki>}</nowiki>, |
||
abstract = <nowiki>{</nowiki>Integrating mathematical morphology operations within deep |
abstract = <nowiki>{</nowiki>Integrating mathematical morphology operations within deep |
||
neural networks has been subject to increasing attention |
neural networks has been subject to increasing attention |
||
Line 44: | Line 46: | ||
further implementations within deep convolutional neural |
further implementations within deep convolutional neural |
||
network architectures.<nowiki>}</nowiki>, |
network architectures.<nowiki>}</nowiki>, |
||
− | + | doi = <nowiki>{</nowiki>10.1007/978-3-030-76657-3_34<nowiki>}</nowiki> |
|
<nowiki>}</nowiki> |
<nowiki>}</nowiki> |
||
Revision as of 20:01, 21 May 2021
- Authors
- Alexandre Kirszenberg, Guillaume Tochon, Élodie Puybareau, Jesus Angulo
- Where
- IAPR International Conference on Discrete Geometry and Mathematical Morphology (DGMM)
- Place
- Uppsala, Sweden
- Type
- inproceedings
- Publisher
- Springer
- Keywords
- Image
- Date
- 2021-02-16
Abstract
Integrating mathematical morphology operations within deep neural networks has been subject to increasing attention lately. However, replacing standard convolution layers with erosions or dilations is particularly challenging because the min and max operations are not differentiable. Relying on the asymptotic behavior of the counter-harmonic mean, p-convolutional layers were proposed as a possible workaround to this issue since they can perform pseudo-dilation or pseudo-erosion operations (depending on the value of their inner parameter p), and very promising results were reported. In this work, we present two new morphological layers based on the same principle as the p-convolutional layer while circumventing its principal drawbacks, and demonstrate their potential interest in further implementations within deep convolutional neural network architectures.
Documents
Bibtex (lrde.bib)
@InProceedings{ kirszenberg.21.dgmm, author = {Alexandre Kirszenberg and Guillaume Tochon and \'{E}lodie Puybareau and Jesus Angulo}, title = {Going beyond p-convolutions to learn grayscale morphological operators}, booktitle = {IAPR International Conference on Discrete Geometry and Mathematical Morphology (DGMM)}, year = {2021}, series = {Lecture Notes in Computer Science}, month = may, address = {Uppsala, Sweden}, publisher = {Springer}, pages = {470--482}, abstract = {Integrating mathematical morphology operations within deep neural networks has been subject to increasing attention lately. However, replacing standard convolution layers with erosions or dilations is particularly challenging because the min and max operations are not differentiable. Relying on the asymptotic behavior of the counter-harmonic mean, p-convolutional layers were proposed as a possible workaround to this issue since they can perform pseudo-dilation or pseudo-erosion operations (depending on the value of their inner parameter p), and very promising results were reported. In this work, we present two new morphological layers based on the same principle as the p-convolutional layer while circumventing its principal drawbacks, and demonstrate their potential interest in further implementations within deep convolutional neural network architectures.}, doi = {10.1007/978-3-030-76657-3_34} }