Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Sort-free Gaussian Splatting via Weighted Sum Rendering
Authors: Qiqi Hou, Randall Rauwendaal, Zifeng Li, Hoang Le, Farzad Farhadzadeh, Fatih Porikli, Alexei Bourd, Amir Said
ICLR 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results show that optimizing a generalized Gaussian splatting formulation to the new differentiable rendering yields competitive image quality. The method was implemented and tested in a mobile device GPU, achieving on average 1.23 faster rendering. ... Table 1: PSNR scores of our method on the Mip-Ne RF360 dataset, the Tanks & Temples dataset, and the Deep Blending dataset. Our method achieved comparable results with 3DGS. |
| Researcher Affiliation | Industry | Qiqi Hou1, Randall Rauwendaal2, Zifeng Li2, Hoang Le1, Farzad Farhadzadeh1, Fatih Porikli1, Alexei Bourd2, Amir Said 1 1Qualcomm AI Research 2Graphics Research Team 1EMAIL |
| Pseudocode | No | The paper describes methods using prose and mathematical equations but does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper states, "We implemented our sort-free Gaussian Splatting method using Py Torch" and mentions "Please refer to our supplementary for more details about our on mobile implementation." However, it does not explicitly state that the code is being released, nor does it provide a direct link to a code repository. |
| Open Datasets | Yes | To ensure a fair comparison, we followed the evaluation setting of 3DGS Kerbl et al. (2023) and conducted our experiments on 13 real-world scenes. Specifically, they include the complete set of scenes from the Mip-Nerf360 dataset Barron et al. (2022), two scenes from the Tanks & Temples dataset Knapitsch et al. (2017), and two scenes from the Deep Blending dataset Hedman et al. (2018). |
| Dataset Splits | No | The paper mentions using well-known datasets and following the evaluation setting of 3DGS Kerbl et al. (2023) but does not explicitly detail the training, validation, or test splits used for its own experiments in the main text. |
| Hardware Specification | Yes | The method was implemented and tested in a mobile device GPU, achieving on average 1.23 faster rendering. ... We examine the running time and memory consumption of our methods on the mobile devices, using an Snapdragon 8 Gen 3 GPU, and with an implementation based on Vulkan. ... For model efficiency, we report run times on a Qualcomm Adreno TM GPU from Snapdragon 8 gen 3 chipset. |
| Software Dependencies | No | The paper mentions "We trained our sort-free Gaussian Splatting method using Py Torch" and "implemented our method and the competitive 3DGS method in Vulkan 2", and "implemented the custom CUDA kernels", but no specific version numbers are provided for PyTorch, Vulkan, or CUDA. |
| Experiment Setup | Yes | Loss function. We optimize our sort-free Gaussian Splatting using the same rendering loss from 3DGS Kerbl et al. (2023), which contains the ℓ1 loss and D-SSIM loss with a factor of 0.2. ... For EXP-WSR, we initialize the σ = 0.1 and β = 0.8. For LC-WSR, we initialize σ = 10 and vi = 0.1. ... where each image is rendered at a resolution of 1920 1080. ... the render target for both is set to use 16-bit RGBA. |