ImageDream

ImageDream Reconstruction

Peng Wang, Yichun Shi

Project Page | Paper | Demo

imagedream-threestudio-teaser

Installation

This part is the same as original MVDream-threestudio. Skip it if you already have installed the environment.

Quickstart

Clone the modelcard on the Huggingface ImageDream Model Page under ./extern/ImageDream/release_models/

In the paper, we use the configuration with soft-shading. It would need an A100 GPU in most cases to compute normal:

…

For diffusion only model, refer to subdir ./extern/ImageDream/ Check ./threestudio/scripts/run_imagedream.sh for a bash example.

Credits

• This code is forked from threestudio and MVDream for SDS and 3D Generation.

Tips

1. Place the object in the center and do not make it too large/small in the image.
2. If you have an object cutting image edge, in config, tuning the parameters range of elevation and fov to be a larger range, e.g. [0, 30], otherwise, you may do image outpainting and follow tips 1.
3. Check the results with ImageDream diffusion model before using it in 3D rendering to save time.

PreComputed Results

• Since there is some randomness in diffusion model and time costly to get baseline results. We put our pre-computed results for reproducing Tab.1 in the paper in a hugging face dataset card

Citing

If you find ImageDream helpful, please consider citing:

@article{wang2023imagedream,
  title={ImageDream: Image-Prompt Multi-view Diffusion for 3D Generation},
  author={Wang, Peng and Shi, Yichun},
  journal={arXiv preprint arXiv:2312.02201},
  year={2023}
}