Is square matrix-shaped kernels provide some advantages than others?
Can we use other shapes like a rectangular matrix for CNNs in some situations?
The square-shaped kernels are preferred because the pattern could occur both horizontally or vertically and by using square-shaped filtered both patterns can be extracted, this sort of ensures symmetry. In some cases with prior assumptions and knowledge of the geometry of the image nonsquare kernel can also be used.
The pattern might appear both horizontally or vertically and both patterns can
be derived by using square-shaped filtering patterns, this sort of ensures symmetry.
Hi Ashay Singh
You can go with these links that can help you up to a certain extent
Deleted research item The research item mentioned here has been deleted
Article Design of a histopathology image segmentation algorithm for ...
Best wishes...!!!!!!!
Ashay Singh , Please refer to these articles,
"Indexed operations for non-rectangular lattices applied to convolutional neural networks", Jacquemont, Mikae and Antiga, Luca and Vuillaume, Thomas and Silvestri, Giorgia and Benoit, Alexandre and Lambert, Patrick and Maurin, Gilles.
"4d spatio-temporal convnets: Minkowski convolutional neural networks", Choy, Christopher and Gwak, JunYoung and Savarese, Silvio, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2019.
Square kernels are used as we don't have any preference in looking for patterns in horizontal and vertical directions. That is why the network is kept symmetrical while looking for features.
The use of asymmetric kernels can be found in inception nets where n*n kernel can be thought as n*1 and 1*n kernels. In effect it has the same receptive field as n*n kernel but utrilizes less no of parameters thus being more efficient.
Square shape filters ensures symmetry and completeness in searching the patterns
In comparison to rectangular matrices, square matrices have very beautiful computational properties., e.g., symmetry, closure properties w.r.t multiplication, addition etc. See e.g., Article Matrix Comparison, Goodness-of-Fit, and Spatial Interaction Modeling
https://online.stat.psu.edu/stat505/lesson/12/12.9
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing
- Image Processing