OMG: Opacity Matters in Material Modeling with Gaussian Splatting
Authors: Silong Yong, Venkata Nagarjun Pudureddiyur Manivannan, Bernhard Kerbl, Zifu Wan, Simon Stepputtis, Katia Sycara, Yaqi Xie
ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We implement our method into 3 different baselines that use Gaussian Splatting for inverse rendering and achieve significant improvements universally in terms of novel view synthesis and material modeling. Code is available at https://github.com/Silong Yong/OMG (...) To verify the correctness and effectiveness of the proposed formulation, we analytically conduct Taylor expansion to our approach and compare the difference w.r.t. the original way of computing opacity. Empirically, we apply the modification to 3 state of the art baselines, namely Gaussian Shader (Jiang et al., 2024), GS-IR (Liang et al., 2024) and R3DG (Gao et al., 2023) and conduct experiments on both synthetic and real-world data. The experimental results on Synthetic4Relight (Zhang et al., 2022), Shiny Blender (Verbin et al., 2022), Glossy Synthetic (Liu et al., 2023) and MIP-Ne RF 360 (Barron et al., 2022) show that our approach enables universal performance improvement in terms of across different baselines and different data. The improvements on material modeling, i.e., albedo estimation and roughness estimation, lead to better novel view synthesis and relighting results in terms of PSNR, SSIM and LPIPS. |
| Researcher Affiliation | Academia | Silong Yong Venkata Nagarjun Pudureddiyur Manivannan Bernhard Kerbl Zifu Wan Simon Stepputtis Katia Sycara Yaqi Xie Carnegie Mellon University EMAIL |
| Pseudocode | No | The paper describes the proposed method and analysis using text and mathematical equations in sections such as 3 PRELIMINARIES and 4 METHOD, but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code is available at https://github.com/Silong Yong/OMG |
| Open Datasets | Yes | The experimental results on Synthetic4Relight (Zhang et al., 2022), Shiny Blender (Verbin et al., 2022), Glossy Synthetic (Liu et et al., 2023) and MIP-Ne RF 360 (Barron et al., 2022) show that our approach enables universal performance improvement in terms of across different baselines and different data. |
| Dataset Splits | No | The paper mentions evaluating on various datasets (Shiny Blender, Glossy Synthetic, Synthetic4Relight, MIP-Ne RF 360) and for Gaussian Shader, it states 'as in the orignal work' for its evaluation settings. However, it does not provide specific details such as exact percentages or sample counts for training, validation, and test splits for any of the datasets in the main text. |
| Hardware Specification | Yes | All the experiments are conducted on a single NVIDIA RTX 6000 Ada Generation GPU and we report the reproduced baseline results for fair comparison. |
| Software Dependencies | No | The paper describes the use of an MLP with ReLU activation and sigmoid output, but does not specify any programming languages, libraries, or solvers with version numbers that would be required to replicate the experiments. |
| Experiment Setup | Yes | The neural network introduced uses 0.001, 0.0001, 0.0005 and 0.007 as learning rate respectively for Synthetic4Relight, Shiny Blender, Glossy Synthetic and MIP-Ne RF 360 datasets.The neural network is implemented as a fully-connected MLP with two hidden layers, 128 as hidden dimension and Re LU as activation function. |