Skip to content

Latest commit

 

History

History
123 lines (81 loc) · 3.9 KB

README.md

File metadata and controls

123 lines (81 loc) · 3.9 KB
Raw

Bootstrap 3D-GS

Overview

In our paper, we present a bootstrapping method that significantly addresses this problem. This approach employs a diffusion model to enhance the rendering of novel views using trained 3D-GS, thereby streamlining the training process.

Our results indicate that bootstrapping effectively reduces artifacts, as well as clear enhancements on the evaluation metrics. Furthermore, we show that our method is versatile and can be easily integrated, allowing various 3D reconstruction projects to benefit from our approach.

Installation

Our program is tested on cuda 11.8 with pytorch==2.0.1, this environment is suitable for both 3D-GS and the stable-diffusion (xformer)

  1. Clone this repo:
git clone https://github.com/yileijin/Bootstrap-3D-GS.git --recursive
  1. Install dependencies
SET DISTUTILS_USE_SDK=1 # Windows only
conda create -n boot_3dgs python==3.11.8
conda activate boot_3dgs
pip install -r requirements.txt

Data

First, create a data/ folder inside the project path by

mkdir data

The data structure will be organised as follows:

data/
├── dataset_name
│   ├── scene1/
│   │   ├── images
│   │   │   ├── IMG_0.jpg
│   │   │   ├── IMG_1.jpg
│   │   │   ├── ...
│   │   ├── sparse/
│   │       └──0/
│   ├── scene2/
│   │   ├── images
│   │   │   ├── IMG_0.jpg
│   │   │   ├── IMG_1.jpg
│   │   │   ├── ...
│   │   ├── sparse/
│   │       └──0/
...

Stable Diffusion CKPT

The stable diffusion model we use can be downloaded in SD2.1v/x4-upscaling

Public Data

The MipNeRF360 scenes are provided by the paper author here. And we test on scenes bicycle, bonsai, counter, garden, kitchen, room, stump. The SfM data sets for Tanks&Temples and Deep Blending are hosted by 3D-Gaussian-Splatting here. Download and uncompress them into the data/ folder.

Training

Our training process is the same as the original 3D-GS, but regarding the Diffusion Model part, we strongly recommend that you make modifications yourself in the configuration ./arguments . Some important configurations are detailed in our paper. For example, to train a scene:

python train.py -s <path to COLMAP or NeRF Synthetic dataset>

Novel-view Rendering

Our rendering also includes scenes that are not in the training set or the test set. By setting do_expension=True, you can obtain completely novel views.

python render.py -m <path to trained model> # Generate renderings
python metrics.py -m <path to trained model> # Compute error metrics on renderings

LICENSE

Please follow the LICENSE of 3D-GS.

Acknowledgement

We thank all authors from 3D-GS for presenting such an excellent work.