DeepDanbooru VS nni

Okay, it turns out that someone has trained an image classifier on danbooru as a dataset.

First of all - DeepDanbooru is the exclusive project of KichangKim, who aimed to train models based on (variants of) the ResNet architecture to output Danbooru tags.

> pip install -r requirements.txt Like what does this mean, where do i put the files i got from https://github.com/KichangKim/DeepDanbooru ? The whole process is confusing me

For: https://github.com/KichangKim/DeepDanbooru I see the "Put your deepbooru release project folder here.txt" file so I "git clone" it into this folder, moved it up and so on but the button did not show and in the console are no errors. Also I restarted the UI after each moving of the project.

if not is_installed("deepdanbooru") and deepdanbooru: run_pip("install git+https://github.com/KichangKim/DeepDanbooru.git@edf73df4cdaeea2cf00e9ac08bd8a9026b7a7b26#egg=deepdanbooru[tensorflow] tensorflow==2.10.0 tensorflow-io==0.27.0", "deepdanbooru")

i think you could make an argument that training an art-generator model falls afoul of the "market for the original work" prong of fair use in a pretty severe way. but something like DeepDanbooru where the output is not the image but a descriptor of it would be a much easier sell as fair use.

DeepDanbooru is suprisingly good for tagging photos and drawings.

For a complete tutorial, navigate to this Jupyter Notebook: https://github.com/microsoft/nni/blob/master/examples/notebooks/tabular_data_classification_in_AML.ipynb

optuna, hyperopt, nni, plenty of less-known tools too.

microsoft / nni