Graph-Guided Scene Reconstruction from Images with 3D Gaussian Splatting
Authors: Chong Cheng, Gaochao Song, Yiyang Yao, Qinzheng Zhou, Gangjian Zhang, Hao Wang
ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate Graph GS achieves high-fidelity 3D reconstruction from images, which presents state-of-the-art performance through quantitative and qualitative evaluation across multiple datasets. The paper includes a dedicated '4 EXPERIMENTS' section, detailing Datasets and Metrics, Implementation, Experimental Results, and Ablation Study, supported by multiple tables (1, 2, 3, 4, 5, 7, 8, 9, 10) and figures (4, 5, 6, 11) for empirical validation. |
| Researcher Affiliation | Academia | The authors are affiliated with HKUST(GZ) (Hong Kong University of Science and Technology (Guangzhou)), HKU (The University of Hong Kong), SCUT (South China University of Technology), and UC Berkeley (University of California, Berkeley), all of which are academic institutions. The affiliation block states: '1HKUST(GZ) 2HKU 3SCUT 4UC Berkeley'. |
| Pseudocode | No | The paper describes its methodology in prose and uses diagrams (Figure 1, 2, 3, 8, 9, 10) to illustrate concepts, but it does not contain any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | The paper provides a project page link: 'Project Page: https://3dagentworld.github.io/ graphgs/.' This is a specific GitHub.io page for the project, implying direct access to project resources including code. |
| Open Datasets | Yes | The paper uses well-known public datasets and cites them: 'We evaluate our method on three datasets: Waymo (Sun et al., 2020), KITTI (Geiger et al., 2012), and Mill-19 (Turki et al., 2022).' |
| Dataset Splits | Yes | The paper specifies a clear dataset split methodology: 'Consistent with prior research, we select one out of every eight images for testing, using the remainder for training.' |
| Hardware Specification | Yes | The paper explicitly describes the hardware used: 'Experiments were conducted using an NVIDIA RTX 3090 GPU and an AMD EPYC 7542 CPU.' |
| Software Dependencies | No | The paper mentions software tools like COLMAP and Open CV but does not specify their version numbers or any other software dependencies with specific version information. |
| Experiment Setup | Yes | The paper provides specific experimental setup details, including hyperparameter values and configurations: 'For the CNNP configuration, the settings of r = 5, h = 20, w = 1 were employed.' and 'To optimize multi-view consistency and balance the coefficients, we set the coefficient λ to 0.07, which was empirically found to be optimal in our experiments.' It also states: 'We set the minimum sampling probability at 0.5 to ensure that nodes with lower weights are not overlooked.' |