SmileSplat: Generalizable Gaussian Splats for Unconstrained Sparse Images

Yanyan Li · Yixin Fang · Federico Tombari · Gim Hee Lee

When I turn to Dust, you will see my Smile (待我成塵時，你將見我微笑) --- Lu Xun (魯迅)

Paper | Project Page | Code (Coming Soon)

Table of Contents

1. Results
2. Installation
3. Pre-trained Checkpoints
4. Public Datasets
5. Running the Code
6. Acknowledgements
7. Citation

Results

Comparison of novel view rendering on the public datasets

Comparison of novel view rendering on the sparse-Replica sequences

Installation

1. Clone NoPoSplat.

2. Create the environment, here we show an example using conda.

Pre-trained Checkpoints

Datasets

Public datasets

Sparse Replica

The Replica dataset is re-created by this paper, which can be download here.

You also can build your own sparse-view dataset via the tool.

Running the Code

Training

Evaluation

Novel View Synthesis

Scaled Gaussian Radiance Field Visualization

Pose Estimation

Acknowledgements

Citation