因为想要泛化模型的适用性，数据集的随机性很大，导致图集太过庞大(12w张图33g)，训练步数仅仅走了5倍（理论训练步数在50-100倍）(5步就已经让我的电脑不眠不休的跑了6天)，导致在某些方面表现的像一坨屎，这个模型就是一坨屎，仅供学习参考。1.非常能出涂鸦效果以及一些崩坏的图。2.迭代步数尽量在50步以上。3.后续版本会精简数据集，可能在此基础上的泛化性会降低。Due to the high randomness of the dataset in order to generalize the applicability of the model, the dataset is too large (12w images, 33g), and the training steps only need to be taken 5 times (theoretical training steps are 50-100 times) (5 steps already make my computer run non-stop for 6 days without sleeping), resulting in the model performing like a piece of shit in some aspects. This model is just a piece of shit, a reference for children's learning.1. Very capable of producing outstanding effects and some damaged images.2. Learn in 50 or more iterations.3. Subsequent versions will have specialized datasets, which may result in reduced generalization on this basis.

Files with SHA256: 4cd315a2d75f4f94ec4696263d5f59108975a63240801efa033520e11b44c7c0