---
license: mit
tags:
- gaussian-splatting
- 3d-gaussian-splatting
- reinforcement-learning
- novel-view-synthesis
---

# RLGS models

Pretrained 3D Gaussian Splatting models and per-scene evaluation results for
**RLGS: Reinforcement Learning-Based Adaptive Hyperparameter Tuning for Gaussian Splatting**.

Paper: [arXiv:2508.04078](https://arxiv.org/abs/2508.04078)

## Repository layout

```
rlgs_tmgs/
  tandt/<scene>/        # Tanks and Temples
  db/<scene>/           # Deep Blending
  dl3dv/<scene>/        # DL3DV
rlgs_mcmc/
  mipnerf360/<scene>/   # Mip-NeRF 360
```

Each scene folder contains:

| File | Contents |
|---|---|
| `point_cloud/iteration_30000/point_cloud.ply` | Trained Gaussians at iteration 30,000 |
| `results.json` | PSNR, SSIM and LPIPS |

## Download

```bash
hf download stack93/RLGS_models --local-dir RLGS_models --include "rlgs_tmgs/tandt/truck/*"
```

Remove `--include` to download everything (about 44 GB).

## Results

Metrics from each scene's `results.json`, at iteration 30,000.

| Dataset | Folder | Scenes | PSNR (dB) | SSIM | LPIPS |
|---|---|---:|---:|---:|---:|
| Tanks and Temples | `rlgs_tmgs/tandt` | 2 | 24.74 | 0.866 | 0.158 |
| Deep Blending | `rlgs_tmgs/db` | 2 | 30.26 | 0.911 | 0.233 |
| DL3DV | `rlgs_tmgs/dl3dv` | 140 | 29.96 | 0.912 | 0.125 |
| Mip-NeRF 360 | `rlgs_mcmc/mipnerf360` | 7 | 30.12 | 0.897 | 0.144 |

### Tanks and Temples

| Scene | PSNR (dB) | SSIM | LPIPS |
|---|---:|---:|---:|
| train | 23.24 | 0.837 | 0.191 |
| truck | 26.25 | 0.895 | 0.124 |

### Deep Blending

| Scene | PSNR (dB) | SSIM | LPIPS |
|---|---:|---:|---:|
| drjohnson | 29.88 | 0.910 | 0.232 |
| playroom | 30.65 | 0.913 | 0.234 |

### DL3DV

<details>
<summary>Show all 140 scenes</summary>

| Scene | PSNR (dB) | SSIM | LPIPS |
|---|---:|---:|---:|
| 032dee | 31.94 | 0.937 | 0.087 |
| 0569e8 | 29.96 | 0.929 | 0.059 |
| 06da79 | 29.19 | 0.913 | 0.138 |
| 073f5a | 23.96 | 0.737 | 0.266 |
| 07d9f9 | 38.85 | 0.984 | 0.023 |
| 085397 | 35.49 | 0.963 | 0.076 |
| 093ef3 | 31.91 | 0.956 | 0.097 |
| 0a1b7c | 33.87 | 0.967 | 0.077 |
| 0a4853 | 33.25 | 0.956 | 0.076 |
| 0bfdd0 | 25.94 | 0.808 | 0.208 |
| 119fd5 | 27.47 | 0.866 | 0.218 |
| 126493 | 24.25 | 0.760 | 0.218 |
| 14eb48 | 29.28 | 0.898 | 0.185 |
| 15ff83 | 27.89 | 0.871 | 0.132 |
| 165f5a | 25.65 | 0.813 | 0.178 |
| 183dd2 | 31.34 | 0.939 | 0.126 |
| 1ba74c | 32.94 | 0.934 | 0.169 |
| 1d6a9e | 23.23 | 0.780 | 0.183 |
| 1da888 | 33.85 | 0.958 | 0.095 |
| 1de58b | 37.46 | 0.980 | 0.037 |
| 238554 | 29.46 | 0.932 | 0.083 |
| 26fd23 | 30.38 | 0.908 | 0.114 |
| 280abf | 32.10 | 0.953 | 0.085 |
| 286239 | 26.01 | 0.866 | 0.126 |
| 2991a7 | 26.30 | 0.854 | 0.164 |
| 2b65ba | 28.30 | 0.913 | 0.120 |
| 2beaca | 32.86 | 0.956 | 0.066 |
| 2cbfe2 | 35.65 | 0.974 | 0.045 |
| 2f3e1c | 28.21 | 0.888 | 0.149 |
| 32c2b9 | 32.75 | 0.927 | 0.205 |
| 341b4f | 28.93 | 0.943 | 0.096 |
| 35317e | 27.92 | 0.887 | 0.117 |
| 358723 | 33.51 | 0.964 | 0.051 |
| 374ffd | 33.58 | 0.952 | 0.070 |
| 387eeb | 32.95 | 0.953 | 0.081 |
| 389a46 | 32.35 | 0.941 | 0.186 |
| 3b16a1 | 26.19 | 0.860 | 0.155 |
| 3b7529 | 32.44 | 0.950 | 0.086 |
| 3bb3bb | 30.50 | 0.946 | 0.111 |
| 3bb894 | 28.44 | 0.904 | 0.105 |
| 410367 | 32.89 | 0.958 | 0.078 |
| 444da1 | 28.05 | 0.932 | 0.116 |
| 457e9a | 29.02 | 0.919 | 0.188 |
| 484c0a | 27.91 | 0.920 | 0.154 |
| 493816 | 28.96 | 0.928 | 0.065 |
| 4ae797 | 30.28 | 0.940 | 0.088 |
| 4ff865 | 34.09 | 0.965 | 0.063 |
| 50c46c | 32.56 | 0.947 | 0.135 |
| 513e4e | 30.57 | 0.932 | 0.116 |
| 54bf35 | 28.84 | 0.905 | 0.117 |
| 56452d | 25.19 | 0.792 | 0.230 |
| 565553 | 29.76 | 0.929 | 0.084 |
| 599ca3 | 27.42 | 0.895 | 0.160 |
| 5a27c0 | 31.55 | 0.960 | 0.063 |
| 5a69d1 | 26.01 | 0.844 | 0.133 |
| 5c3af5 | 31.73 | 0.952 | 0.053 |
| 5c8daf | 31.52 | 0.955 | 0.098 |
| 5f0041 | 33.94 | 0.969 | 0.046 |
| 63798f | 32.20 | 0.941 | 0.187 |
| 669c36 | 28.80 | 0.933 | 0.111 |
| 66fd66 | 33.53 | 0.965 | 0.082 |
| 6d2216 | 27.62 | 0.862 | 0.225 |
| 6d81c5 | 24.80 | 0.829 | 0.230 |
| 6e11e7 | 34.06 | 0.959 | 0.165 |
| 70eac6 | 27.53 | 0.922 | 0.104 |
| 71b2dc | 35.46 | 0.974 | 0.063 |
| 75fbbe | 30.82 | 0.939 | 0.099 |
| 7705a2 | 31.48 | 0.908 | 0.189 |
| 7a9f97 | 31.00 | 0.926 | 0.116 |
| 7da3db | 29.14 | 0.917 | 0.148 |
| 800cf8 | 27.47 | 0.870 | 0.146 |
| 8324b3 | 25.88 | 0.887 | 0.139 |
| 85cd0e | 29.11 | 0.906 | 0.145 |
| 8b9fb9 | 24.44 | 0.825 | 0.172 |
| 8cb2e9 | 31.67 | 0.924 | 0.155 |
| 8fdc51 | 29.46 | 0.913 | 0.125 |
| 90cb7e | 30.15 | 0.938 | 0.091 |
| 917e9c | 32.14 | 0.955 | 0.077 |
| 918c8d | 27.61 | 0.892 | 0.138 |
| 91afb9 | 32.66 | 0.956 | 0.076 |
| 946f49 | 33.05 | 0.952 | 0.065 |
| 9641a1 | 26.71 | 0.856 | 0.237 |
| 9c8c0e | 25.97 | 0.821 | 0.155 |
| 9cbc55 | 30.85 | 0.945 | 0.078 |
| 9e9a89 | 31.97 | 0.950 | 0.156 |
| 9fb058 | 28.37 | 0.928 | 0.087 |
| a17a98 | 31.20 | 0.952 | 0.094 |
| a40146 | 28.09 | 0.898 | 0.122 |
| a62c33 | 27.15 | 0.894 | 0.123 |
| a62f9a | 32.26 | 0.955 | 0.087 |
| a726c1 | 26.24 | 0.899 | 0.191 |
| adb95f | 27.01 | 0.842 | 0.181 |
| adf351 | 30.42 | 0.906 | 0.204 |
| af0d70 | 27.37 | 0.906 | 0.124 |
| b2076b | 30.02 | 0.944 | 0.104 |
| b3bf90 | 30.38 | 0.923 | 0.099 |
| b4f530 | 31.42 | 0.951 | 0.063 |
| b5faa2 | 28.65 | 0.852 | 0.238 |
| b6d113 | 30.75 | 0.951 | 0.060 |
| b92b49 | 33.79 | 0.962 | 0.106 |
| ba55c8 | 35.61 | 0.964 | 0.073 |
| c07692 | 38.05 | 0.980 | 0.052 |
| c37109 | 22.17 | 0.686 | 0.292 |
| c37726 | 25.54 | 0.839 | 0.193 |
| c45589 | 27.41 | 0.890 | 0.135 |
| cbd44b | 34.19 | 0.962 | 0.083 |
| cc08c0 | 28.47 | 0.900 | 0.141 |
| cd9c98 | 28.76 | 0.924 | 0.095 |
| ceb252 | 33.90 | 0.958 | 0.084 |
| d1b3a0 | 25.80 | 0.867 | 0.135 |
| d3812a | 21.68 | 0.756 | 0.316 |
| d3af82 | 32.58 | 0.942 | 0.091 |
| d4fbeb | 26.72 | 0.893 | 0.094 |
| d8de66 | 29.42 | 0.897 | 0.144 |
| d904ae | 25.91 | 0.835 | 0.171 |
| d9b637 | 24.59 | 0.792 | 0.202 |
| d9f4c7 | 34.92 | 0.975 | 0.047 |
| dac979 | 31.13 | 0.949 | 0.116 |
| dafa9c | 23.86 | 0.811 | 0.170 |
| ddfdcf | 29.41 | 0.938 | 0.080 |
| ded5e4 | 32.58 | 0.930 | 0.166 |
| df04f5 | 33.11 | 0.960 | 0.042 |
| df29c2 | 35.71 | 0.964 | 0.061 |
| df4f9d | 33.27 | 0.934 | 0.197 |
| e5684b | 29.58 | 0.906 | 0.171 |
| e78f8c | 30.91 | 0.945 | 0.054 |
| e8ce51 | 27.98 | 0.932 | 0.124 |
| e9360e | 26.19 | 0.885 | 0.146 |
| eb4cf5 | 31.26 | 0.937 | 0.148 |
| ec1e44 | 29.71 | 0.938 | 0.123 |
| ec3057 | 29.22 | 0.926 | 0.092 |
| ed1632 | 29.21 | 0.891 | 0.118 |
| ef59aa | 29.83 | 0.866 | 0.139 |
| f004c8 | 35.13 | 0.976 | 0.055 |
| f477ff | 31.14 | 0.946 | 0.153 |
| f71ac3 | 30.04 | 0.936 | 0.053 |
| f7aaea | 28.44 | 0.929 | 0.086 |
| fb2c04 | 28.12 | 0.909 | 0.110 |
| fb3b73 | 24.34 | 0.773 | 0.210 |
| ff5923 | 33.09 | 0.948 | 0.108 |

</details>

### Mip-NeRF 360

| Scene | PSNR (dB) | SSIM | LPIPS |
|---|---:|---:|---:|
| bicycle | 26.30 | 0.810 | 0.163 |
| bonsai | 33.26 | 0.955 | 0.160 |
| counter | 29.70 | 0.927 | 0.159 |
| garden | 28.36 | 0.887 | 0.087 |
| kitchen | 32.64 | 0.941 | 0.105 |
| room | 32.99 | 0.940 | 0.168 |
| stump | 27.60 | 0.818 | 0.166 |

## License

This repository is released under the MIT License.

## Citation

```bibtex
@article{li2025rlgs,
  title={RLGS: Reinforcement Learning-Based Adaptive Hyperparameter Tuning for Gaussian Splatting},
  author={Li, Zhan and Zhan, Huangying and Li, Changyang and Yan, Qingan and Xu, Yi},
  journal={arXiv preprint arXiv:2508.04078},
  year={2025}
}
```