pollinations / adampi

Create a 3D photo from single in-the-wild 2D images

  • Public
  • 5.7K runs
  • GitHub
  • Paper
  • License
Run with an API
Playground API Examples README Versions

😵 Uh oh! This model can't be run on Replicate because it was built with a version of Cog or Python that is no longer supported. Consider opening an issue on the model's GitHub repository to see if it can be updated to use a recent version of Cog. If you need any help, please Contact us about it.

Examples

View more

Run time and cost

This model costs approximately $0.027 to run on Replicate, or 37 runs per $1, but this varies depending on your inputs. It is also open source and you can run it on your own computer with Docker.

This model runs on Nvidia T4 GPU hardware. Predictions typically complete within 119 seconds. The predict time for this model varies significantly based on the inputs.

Readme

Forked to be able to run as a COG ,odel

News

  • 2022/8/4 We release the code and the pretrained model for AdaMPI. Use it to create 3D photo from single in-the-wild 2D images!
  • 2022/7/22 We release the code and the pretrained inpainting model for the Warp-Back strategy.

AdaMPI

This is a PyTorch implementation of the following paper:

Single-View View Synthesis in the Wild with Learned Adaptive Multiplane Images, SIGGRAPH 2022.

Yuxuan Han, Ruicheng Wang and Jiaolong Yang

Project Page | Video | Paper

Abstract: This paper deals with the challenging task of synthesizing novel views for in-the-wild photographs. Existing methods have shown promising results leveraging monocular depth estimation and color inpainting with layered depth representations. However, these methods still have limited capability to handle scenes with complex 3D geometry. We propose a new method based on the multiplane image (MPI) representation. To accommodate diverse scene layouts in the wild and tackle the difficulty in producing high-dimensional MPI contents, we design a network structure that consists of two novel modules, one for plane depth adjustment and another for depth-aware color prediction. The former adjusts the initial plane positions using the RGBD context feature and an attention mechanism. Given adjusted depth values, the latter predicts the color and density for each plane separately with proper inter-plane interactions achieved via a feature masking strategy. To train our method, we construct large-scale stereo training data using only unconstrained single-view image collections by a simple yet effective warp-back strategy.

Document

This repo contain two parts: * The test code for the AdaMPI pretrained model, you can use it to generate 3D photo from single-view in-the-wild image input, as shown in the teaser gif. See AdaMPI.md for more details.

  • The implementation of the Warp-Back strategy, you can use it to generate multi-view training data to train your own view synthesis model (e.g. SynSin, WorldSheet) using only single-view image collections like COCO dataset. See Warp-Back.md for more details.

Contact

If you have any questions, please contact Yuxuan Han (hanyuxuan076@gmail.com).

License and Citation

This repository can only be used for personal/research/non-commercial purposes. Please cite the following paper if this model helps your research:

@inproceedings{han2022single,
    author = {Han, Yuxuan and Wang, Ruicheng and Yang, Jiaolong},
    title = {Single-View View Synthesis in the Wild with Learned Adaptive Multiplane Images},
    booktitle = {ACM SIGGRAPH},
    year={2022}
}

Acknowledgments

  • The code of MPI rendering operation (utils/mpi/*.py) is heavily borrowed from MINE, we thank the author for their great effort.
  • We thank Varun Jampani and Meng-Li Shih for helpful suggestions and discussions to evaluate their method.