The CLI arguments '-t', '-v' and '-T' represents amounts of raw images. While it does not really matter for 'shapes' and 'mapillary' datasets, it clarifies the way geographic datasets ('aerial' and 'tanzania') are considered, especially in training and validation scopes.
As a side effect, one needs an additional arguments, 'nb-tiles-per-image', that depicts the amount of tiles that will be drawn from the original raw images (training and validation context). Example: if '-t=10' and '--nb-tiles-per-image=10', one will pick 10 images within the original training set, and try to generate 10 tiles from each of them (same random process as before).
The documentation for preprocessing has been clarified in this way.
obrix
commented
4 years ago
The CLI arguments '-t', '-v' and '-T' represents amounts of raw images. While it does not really matter for 'shapes' and 'mapillary' datasets, it clarifies the way geographic datasets ('aerial' and 'tanzania') are considered, especially in training and validation scopes.
As a side effect, one needs an additional arguments, 'nb-tiles-per-image', that depicts the amount of tiles that will be drawn from the original raw images (training and validation context). Example: if '-t=10' and '--nb-tiles-per-image=10', one will pick 10 images within the original training set, and try to generate 10 tiles from each of them (same random process as before).
The documentation for preprocessing has been clarified in this way.
As a side topic, this PR updates some of the project dependencies; the identified vulnerabilities should be fixed (https://github.com/Oslandia/deeposlandia/network/alerts).
Linked issue: #147