Fake3DGS

This is the official PyTorch implementation of the paper "Fake3DGS: A Benchmark for 3D Manipulation Detection in Neural Rendering" (ICPR 2026).

1. Setup

Environment Setup

You can find all the packages and dependencies in the environment.yml file. If you have conda, you can simply run

conda env create -f environment.yml

2. Download the dataset

Download the dataset: original_samples | edited_samples

The dataset contains nerfstudio checkpoints compressed. You need to preprocess the data with extract_data.py file to extract each gaussian splats features inside a pkl file, according to our experimentation.

python extract_data.py --input_dir root/Fake3DGS/original_compressed/ --output_dir gaussian_pickels/real --label 0 #1 if fake

3. Training

python -u train.py -bs 4 -e 8 --config mixed_training.pkl --name new_mixed --resume new_mixed_5.pth

Explanation of Parameters:

• --bs : Batch size.
• -e : Number of epochs.
• --config: config pkl file.
• --name: Name of the training.
• --resume: Checkpoint from which to resume training.

4. Testing

To test the model, download the corresponding pretrained weights HERE and place them in the folder:

checkpoints/

Then run:

python -u test.py -bs 4 --test_set "mixed_training.pkl" --ckpt "checkpoints/best.pth"

Explanation of Parameters:

• --bs : Batch size.
• --ckpt : Model weights to load.
• --test_set: config pkl file.

Citation

If you find our work useful for your project, please consider citing the paper:

@inproceedings{ddinuccifake3dgs,
  title={Fake3DGS: A Benchmark for 3D Manipulation Detection in Neural Rendering},
  author={Di Nucci, Davide and Catalini, Riccardo and Borghi, Guido and Vezzani, Roberto},
  booktitle={Twentyeighth International Conference on Pattern Recognition},
  year=2026
}

---