adirik/dreamgaussian

DreamGaussian: Generative Gaussian Splatting for Efficient 3D Content Creation

  • Public
  • 6.5K runs

Run time and cost

This model runs on Nvidia A40 GPU hardware. Predictions typically complete within 5 minutes. The predict time for this model varies significantly based on the inputs.

Readme

DreamGaussian

DreamGaussian is a diffusion-based Image-to-3D and Text-to-3D model that uses Gaussian Splatting to output textured meshes in 2 minutes. See the original repository and paper for details.

DreamGaussian converts the image or text input to a 3D mesh in 2 stages - (i) raw mesh optimization and (ii) refined texture optimization.

Using the Model

To use the model, either upload an image or enter a text prompt to generate a 3D object. If uploading an image, make sure that the image is not overcrowded and the target object is clearly visible in the foreground. DreamGaussian comes with a background removal module that removes the background for you.

Other settings are:
- elevation: The z-axis angle between the camera and target object in the image. Default value is 0, meaning the object is directly in front of the camera at the same height. Set elevation to lower angles (e.g -30) if the camera is above the object and higher angles if the camera is set lower than the object in the image.
- num_steps: Number of steps for the raw optimization stage.
- num_refinement_steps: Number of steps for the refined optimization stage.
- num_point_samples: Number of points sampled from the mesh surface for Gaussian Splatting.

References

@article{tang2023dreamgaussian,
  title={DreamGaussian: Generative Gaussian Splatting for Efficient 3D Content Creation},
  author={Tang, Jiaxiang and Ren, Jiawei and Zhou, Hang and Liu, Ziwei and Zeng, Gang},
  journal={arXiv preprint arXiv:2309.16653},
  year={2023}
}