Perceptual-GS: Scene-adaptive Perceptual Densification for Gaussian Splatting
Authors: Hongbi Zhou, Zhangkai Ni
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive evaluations on multiple datasets, including Bungee Ne RF for large-scale scenes, demonstrate that Perceptual GS achieves state-of-the-art performance in reconstruction quality, efficiency, and robustness. The code is publicly available at: https:// github.com/eezkni/Perceptual-GS |
| Researcher Affiliation | Academia | 1School of Computer Science and Technology, Tongji University, Shanghai, China. Correspondence to: Zhangkai Ni <EMAIL>. |
| Pseudocode | No | The paper describes its methodology using prose, mathematical equations, and diagrams. However, there are no clearly labeled pseudocode or algorithm blocks present in the document. |
| Open Source Code | Yes | The code is publicly available at: https:// github.com/eezkni/Perceptual-GS |
| Open Datasets | Yes | We evaluated the effectiveness of our method across 21 scenes, including 9 scenes from Mip-Ne RF 360 (Barron et al., 2022), 2 scenes from Deep Blending (Hedman et al., 2018), 2 scenes from Tanks & Temples (Knapitsch et al., 2017), and 8 scenes from Bungee Ne RF (Xiangli et al., 2022). |
| Dataset Splits | No | The paper mentions using several datasets for evaluation but does not specify the training, validation, or test splits (e.g., percentages or sample counts) for any of them. It refers to 'training and testing' but without explicit details on how the datasets were partitioned. |
| Hardware Specification | Yes | All training and testing are conducted on a single NVIDIA RTX4090 GPU with 24GB of memory. |
| Software Dependencies | No | The paper mentions various techniques and models like '3DGS' and 'Ne RF' but does not specify any particular software libraries, frameworks, or their version numbers that were used for implementation. |
| Experiment Setup | Yes | We adopt the default settings of 3DGS and show the additional hyperparameters introduced in Perceptual-GS in Table 10. Table 10. Definition and value of hyperparameters introduced in Perceptual-GS. H.P. Definition value τe perception-oriented enhancement threshold 0.05 τs perception-oriented smoothing threshold 0.3 λS sensitivity loss weight 0.1 Iterh high-sensitivity Gaussians densification interval 1000 Iterm medium-sensitivity Gaussians densification interval 1500 τh high-sensitivity Gaussians threshold of perceptual sensitivity 0.9 τl low-sensitivity Gaussians threshold of perceptual sensitivity 0.3 τ ω h high-sensitivity Gaussians threshold of weight 25 τ ω m medium-sensitivity Gaussians threshold of weight 10 τβ high-sensitivity scenes threshold 0.85 τγ scenes with sparse initial point cloud threshold 0.55 |