GNeRP: Gaussian-guided Neural Reconstruction of Reflective Objects with Noisy Polarization Priors
Authors: LI Yang, RUIZHENG WU, Jiyong Li, Ying-Cong Chen
ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To validate the effectiveness of our design, we capture polarimetric information, and ground truth meshes in additional reflective scenes with various geometry. We also evaluated our framework on the PANDORA dataset. Comparisons prove our method outperforms existing neural 3D reconstruction methods in reflective scenes by a large margin. |
| Researcher Affiliation | Collaboration | Yang LI1,2,4 Ruizheng WU4 Jiyong LI3 Yingcong CHEN1,2, 1AI Thrust, HKUST(GZ), Nansha, Guangzhou, China 2Department of Computer Science & Engineering, HKUST, Clear Water Bay, Hong Kong SAR, China 3Department of Computer Science, Sun Yat-sen University, Panyu, Guangzhou, China 4R & D Center, Smart More, Qianhai, Shenzhen, China |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide an explicit statement about releasing the source code for the methodology or a link to a code repository. |
| Open Datasets | No | To comprehensively evaluate the performance of 3D reconstruction methods, a new challenging multi-view dataset named Pol Ref was collected... The dataset will be released to facilitate further research on 3D reconstruction in more challenging scenes in the future. |
| Dataset Splits | No | The paper does not explicitly provide specific training/validation/test dataset splits (e.g., exact percentages or sample counts) needed to reproduce the experiment. |
| Hardware Specification | Yes | The model is trained for 200k iterations and takes about 6 hours on a server with 4 NVIDIA RTX 3090 Ti GPUs for the reconstruction. |
| Software Dependencies | No | The paper mentions 'GNe RP is built upon Neu S' and 'Mitsuba renderer' but does not provide specific version numbers for these or other software dependencies. |
| Experiment Setup | Yes | The model is trained for 200k iterations... The meshes are extracted from learned SDF by Marching Cubes... with a resolution of 5123. The hyper-parameter settings are shown in the Appendix D.3. ... γ and δ are fixed at 0.1 to follow previous methods. β and β are fixed at 0.1. However, α is set to either 0.1 or 1, depending on the overall ratio of reflection regions. |