720.10 b

Description: ChatGPTGaussianDreamer is a rapid 3D asset generation framework that blends 2D and 3D diffusion models, leveraging Gaussian splatting to enrich geometry and appearance. It creates high-quality 3D instances from text prompts within 15 minutes on a single GPU, surpassing previous methods in speed. The generated 3D models can be rendered in real-time..

#700 #Research_Paper#720 #Computer_Vision#720.10 #Text_to_3D#720.10 b #GaussianDreamer:_Fast_Generation_from_Text_to_3D_Gaussians_by_Bridging_2D_and_3D_Diffusion_Models

Output: 3D
Type: .ply
Input: Prompt
Code: https://github.com/hustvl/GaussianDreamer?tab=readme-ov-file
Demo: https://colab.research.google.com/github/taoranyi/GaussianDreamer-colab/blob/main/GaussianDreamer_colab.ipynb
논문: https://arxiv.org/pdf/2310.08529

GPU A40기준 13분 30초 가량 소요

확실히 시간을 많이 들여서 학습하고, 이미지를 자체생성한거 가지고 학습하는 만큼 퀄리티가 좋은느낌.

1200학습으로 프롬프트에서 이미지 및 텍스쳐를 생성하고, 이후 해당 이미지를 120장 (360도방향)을 생성해서 전체적인 모습을 만듬.. 학습이 진행될수록 텍스쳐가 크게 변하고, 메쉬가 세밀하게 변하는게 보임