Open sr229 opened 4 years ago
Firstly, AI is not all-powerful since there will always be errors. Secondly, are you only classifying or are you doing Semantic Segmentation too?
In the beginning, we should check https://github.com/KichangKim/DeepDanbooru
We're going for classification only, the primary aim here is to have grouping by character in the near future - ala-Pixiv
Similar to #8. We can use an AI to train multiple models of anime characters we know and love. There are 4 different ways, for example, to draw Mash Kyrielight from Fate, from four different artists. We should be able to compensate for that. Once we know where the character came from, we can also output the franchise/series that they came from.