A simple implementation of a way to get a field of points from a single photo.
This notebook was used for various tests in the development of a solution for predicting the results of plastic surgery for breast augmentation. Including to determine the main development vector, as well as the possibility of using non-standard or segenerated images for further training of the main model.
To obtain depth from the base image, in this work it was decided to use midas, because in tests it showed the best results for close-up images.
In view of the task, the resulting field of dots was deliberately not returned to normal colors and was not cleared of "noisy" dots, since this was not necessary.
Since there were not very many images to check, this task was solved quite quickly by hand. However, with a large number of images, it is possible to automate the process by marking a small dataset and teaching a simple detector to find deviations in the desired zones in the depths.
P.S. The repository may be updated.