Code implementation for "Adversarial Generation of Hierarchical Gaussians for 3D Generative Model"


- Environment setting
We build our code on EG3D (https://github.com/NVlabs/eg3d) and 3D-GS (https://github.com/graphdeco-inria/gaussian-splatting).
Please dependencies of those two repository to run the codes.
One different thing is that we use different version for "diff-gaussian-rasterization" that support depth and alpha rendering in the below link.
"https://github.com/ashawkey/diff-gaussian-rasterization"


- Data preprocessing
We exatly follow the data-preprocessing of EG3D, so please refer to them (https://github.com/NVlabs/eg3d).


- Training
We provide the training script in "_scripts/train_{dataset}.sh" file.
Only need to change is the "dataset_path" variable.
Training results are saved at "outdir" directory.

- Evaluation
We provide the evaluation of FID.
Set the path of "network_pkl" in "_scripts/eval.sh" file properly, and run it.

- Video generation
Set the path of "network_pkl" in "_scripts/generate_videos.sh" file properly, and run it.
