Request for PCA Visualization Code and Interactive Query Viewer

[Andiezz]

Hi SceneSplat team,

First of all, thank you for the great work on this project. I’ve been exploring the repository and the results shown in the paper/demo are very impressive.

I wanted to ask whether you could release the code for:

1. The PCA-based visualization pipeline

• The code used to generate the PCA feature visualizations shown in the project/paper.
• Any preprocessing or dimensionality reduction details that are important for reproducing the results.

2. The interactive query viewer

• The viewer/interface used for interactive querying and visualization.
• Instructions for running it locally, if available.

These components would be extremely helpful for better understanding the feature representations and reproducing the qualitative results.

Thanks again for the excellent work!

[unique1i]

Hi @Andiezz, thanks for your interest! I have commited the helper scripts for the PCA visualization and our mini viewer.

For PCA visualization, you can now run it directly after inference with:

python -m tools.lang_inference \
    --config configs/inference/lang-pretrain-pt-v3m1-3dgs.py \
    --checkpoint /path/to/model_best.pth \
    --input-root /path/to/preprocessed_scene \
    --output-dir /path/to/output_dir \
    --pca-vis

Or run PCA visualization separately from a saved feature tensor:

python -m scripts.pca_colorize_features \
    --feature-path /path/to/output_dir/<scene>_feat.pt \
    --input-root /path/to/preprocessed_scene \
    --output-dir /path/to/output_dir

The saved feature tensor is <output-dir>/<scene>_feat.pt with shape (N, 768), aligned with the input Gaussians. PCA visualization writes:

• <scene>_pca_colored.ply: colored point cloud using Gaussian centers
• <scene>_feat_vis_3dgs.ply: renderable 3DGS PLY with PCA feature colors

For the interactive viewer, we added a wrapper for our standalone Mini Viewer repo:

python -m pip install "nerfview>=0.1.3" "splines>=0.3" \
    "git+https://github.com/RunyiYang/Mini_Viewer.git@6c8e5c938844487319a92e19f952e76cd4eba847"

Then precompile the gsplat CUDA extension once:

python -m tools.mini_viewer --precompile-gsplat

After running inference, launch the viewer with:

python -m tools.mini_viewer \
    --input-root /path/to/preprocessed_scene \
    --output-dir /path/to/output_dir

Let me know if this helps.