Point'n Move

July 19, 2025 · View on GitHub

Code for paper Point'n Move: Interactive Scene Object Manipulation on Gaussian Splatting Radiance Fields

Setup Environment

Please refer to Justfile task setup_environment, setup_sa3d_chkpts, objremoval_setup_environment for environment setup:

just setup_environment
just setup_sa3d_chkpts
just objremoval_setup_environment 

Download Dataset & Model

  1. Download Dataset from https://huggingface.co/datasets/jjhuangbu/pnm-data, and put it as gaussian3d/data
  2. Download mipnerf360 dataset 360_v2.zip from https://jonbarron.info/mipnerf360/, and unzip it's content into gaussian3d/data
  3. Download trained models from https://huggingface.co/jjhuangbu/pnm-model, and put it as gaussian3d/output

Your result file tree should look like:

.
├── gaussian3d/
│   ├── data/
│   │   ├── garden
│   │   ├── ... (rest of 360_v2.zip)
│   │   ├── mipnerf360-annotation
│   │   └── ... (rest of pnm-data repo)
│   └── output/
│   │   ├──spinnerf-v2_2
│   │   └── ... (rest of pnm-model repo)
├── ...
├── Justfile
└── README.md

Scripts

# qualitative evaluation on MipNeRF360
python -m saga3d.run_saga3d_all_mipnerf360
# quantitative evaluation on SPIn-NeRF
python -m saga3d.run_saga3d_all_spinnerf && python -m saga3d.run_saga3d_all_spinnerf_compile_result
# edit sample (run mipnerf360 eval first)
python -m saga3d.run_saga3d_all_edit
# ablation studies
python -m saga3d.run_saga3d_all_ablation