Hello,
A data augmentation technique called ''elastic deformation'' is used in the U-Net paper to help the network learn invariance to deformations, without the need to see these transformations in the annotated image corpus.
On the 6th page of the U-Net paper, the following sentences are written:
"We generate smooth deformations using random displacement vectors on a coarse 3 by 3 grid. The displacements are sampled from a Gaussian distribution with 10 pixels standard deviation. Per-pixel displacements are then computed using bicubic interpolation."
What is the basis for assuming sampling displacements of pixels from Gaussian distribution? Is it based on the physics of images? or can better distribution be simulated?
U-Net paper: Article U-Net: Convolutional Networks for Biomedical Image Segmentation
Hello
the choice of gaussian distribution for displacement can be selected for smoothness, local coherence, limited range, and easy to control reasons. But in CNN document it is not explained.
- Badges
- Science topic
- Similar topics
- Filing