FEATURE REQUEST: Some QoL for training parameters, bug in output window

[jtkelm2]

Allow the user to choose a folder containing pre-generated class regularization images in the same way they can choose a folder for the instance images. Upon selection of a folder, automatically apply the proper flags (use_prior_preservation, num_class_images, etc.).

• This beats the current method of having to specify the directory and flags manually in the text box. A mistake there can easily lead to class images being generated anyway and having to abort/restart the training session.

Include some hardcoded text fields/drop-down menus for the most common flags.

• I don't want to have to memorize them all according to a list deep within Shivam's repo!

Allow preprocessing of training image directories like resizing, center-cropping, and flipping as in AUTO's webui. Alternatively, just include a checkbox that indicates whether you want to train on a flip-augmented dataset.

For some reason, all the r's have been replaced with n's in the training output window: Maybe a failed attempt at converting carriage returns?

[smy20011]

For the output bug, It should be fixed in 1.4 :)