Closed lkyf closed 11 months ago
I'm not sure if my understanding of diversity is the same as yours: the purpose of training 3D generative models is to try to generate shapes that are consistent with the distribution of the training data. As for the issue of overfitting, you can refer to Fig. 20 in our paper, where we show our model is capable of generating new shapes.
Thank you for your reply. For example, I only trained on hundreds of 3D models of different categories, and the generated models were almost the same as the training data, did too little training data lead to overfitting?
Yes, too little training data would lead to overfitting. Data augmentation may alleviate this problem.
The shape of the generated model is relatively close to the training data. Are there any suggestions to improve the diversity of the generated model?