| Title |
3D Generative Model using Frequency Domain Loss for Effectively Generating Fine-grained Features |
| DOI |
https://doi.org/10.5573/ieie.2024.61.5.52 |
| Keywords |
3D generative model; Frequency domain loss function |
| Abstract |
Existing 3D generative models can generate images that resemble real footage, but they often fail to realistically create fine details such as human hair or ear. We argue that this issue stems from the model's failure to learn features in the frequency domain, resulting in a discrepancy between the frequency spectrum distribution of the generated images and real footage. Therefore, we propose to improve the generation performance of fine details by adding a frequency domain loss function to the 3D generative model, enabling it to learn features in the frequency domain. Through qualitative and quantitative experiments, we observed that our generative model effectively generates fine details realistically and improves even the FID score and IS of image generation. |