Is there technique to create more datasets like data augmentation?. if data augmentation applicable to given datasets then which method of data augmentation helpful.Posted by: anilrai @ April 17, 2020, 7:29 a.m.
+1, most augmentation tricks does not work herePosted by: argman @ April 18, 2020, 1:50 a.m.
I mean , can we create more training patients datasets from given datasets?Posted by: anilrai @ April 18, 2020, 8:17 a.m.
Apart from the usual affine and pixel-wise transformations, you might try elastic and diffeomorphic data augmentations which are suited for medical images. I found a description of these techniques in a nature article which I can't seem to find now. In terms of generating artificial data, you could look into GANs.Posted by: overfitting @ May 13, 2020, 5:03 a.m.
roi_crop, rotation, flip are useful,
scale, elastic deformation do not work