The standalone GUI script provides the same features as the normal GUI. We provide some examples for you to play with:
# Make sure git-lfs is installed (https://git-lfs.com)
git lfs install
# If not, then
sudo apt-get install git-lfs
git clone https://huggingface.co/yunjinli/sadg_example_models
cd sadg_example_models
git lfs pull
python gui_standalone.py -m ./sadg_example_models/sear_steak --iteration 30000
## Or
python gui_standalone.py -m ./sadg_example_models/split-cookie --iteration 30000
As the standalone GUI doesn't have information regarding the length (in second) of the dynamic scene, this needs to be configured per scene in the GUI with the slider Duration (sec) (Default: 10 seconds). For detailed functionalities of the GUI, please see the following sections.
python gui.py -m output/<DATASET>/<NAME> --load_mask_on_the_fly --load_image_on_the_fly --eval --load2gpu_on_the_fly --iteration 30000
- You can use the cursor to drag to different novel views in the GUI
- Scroll to adjust the FoV
- Select different rendering modes (
Render, Rendered Features, Gaussian Features, Gaussian Clusters, Segmentation, Point Cloud, Depth) in the combo box - Please make sure to run
Clusteringbefore rendering the inGaussian Clusters, Segmentationmode
gui_demo_modes.mp4
- Make sure you run
Clusteringbeforehand - Drag the slider of
Freeze Timeto change to different time - Hold the key
AandLeft-Clickon the object of interest in the novel view - Hold the key
DandLeft-Clickto deselect the click - Drag the slider of
Score Thresholdto adjust the threshold for filtering unwanted Gaussians
gui_demo_click_prompt.mp4
- Make sure you run
Clusteringbeforehand - Enter the text prompt and click
Enter - Click
Remove Objectto toggle the visibility of the selected objects (selected / removal)
gui_demo_text_prompt.mp4
- You can change to different clustering methods in the combo box (DBSCAN / K-Means)
Render Maskrenders transparant segmentation mask on the renderingSave Objectsaves the current visible Gaussians (for later composition with other scenes)Render Objectrenders to the test camera views
We appreciate the authors from SC-GS for sharing their amazing work. Our GUI is built upon their code. Please consider also citing their paper.
@article{huang2023sc,
title={SC-GS: Sparse-Controlled Gaussian Splatting for Editable Dynamic Scenes},
author={Huang, Yi-Hua and Sun, Yang-Tian and Yang, Ziyi and Lyu, Xiaoyang and Cao, Yan-Pei and Qi, Xiaojuan},
journal={arXiv preprint arXiv:2312.14937},
year={2023}
}