You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
While the construction of a graph and analysis of its laplacian may contain "sufficient" information for many tasks, it disregards spatial locality which has been an express interest in the past. Instead of going for the raw image-->graph-->laplacian-->spectrum pipeline, we may consider using wavelets to generate a representation, e.g. raw image-->wavelet decomp-->scattering spectrum which has the capacity to preserve local spatial distributions as well. This can be useful for tracking the origin of spatial shifts (if such origins even exist).
The text was updated successfully, but these errors were encountered:
While the construction of a graph and analysis of its laplacian may contain "sufficient" information for many tasks, it disregards spatial locality which has been an express interest in the past. Instead of going for the raw image-->graph-->laplacian-->spectrum pipeline, we may consider using wavelets to generate a representation, e.g. raw image-->wavelet decomp-->scattering spectrum which has the capacity to preserve local spatial distributions as well. This can be useful for tracking the origin of spatial shifts (if such origins even exist).
The text was updated successfully, but these errors were encountered: