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].
HVOFusion: Incremental Mesh Reconstruction Using Hybrid Voxel Octree
Authors: Shaofan Liu, Junbo Chen, Jianke Zhu
IJCAI 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we conduct a series of experiments to evaluate the reconstruction and rendering quality of our proposed approach. Please refer to our supplementary materials for more ablation experiments and visualization results. |
| Researcher Affiliation | Collaboration | Shaofan Liu1 , Junbo Chen2 , Jianke Zhu1 1Zhejiang University 2Udeer.ai |
| Pseudocode | No | The paper describes the method in section 3, but it does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code is available at https://github.com/Frankuzi/HVOFusion. |
| Open Datasets | Yes | We conduct experiments on three different kinds of datasets, including Replica [Straub et al., 2019], Scan Net++ [Yeshwanth et al., 2023], and Newer College Dataset [Ramezani et al., 2020]. |
| Dataset Splits | No | The paper evaluates on 'Novel View' and 'Training View' on Scan Net++ (Table 2), which implies a data partition. However, it does not provide specific details on how these splits are defined or generated (e.g., percentages, sample counts, or explicit splitting methodology) for reproduction purposes beyond using predefined views from the dataset itself. |
| Hardware Specification | Yes | All experiments are conducted on a desktop PC with an NVIDIA RTX3090 GPU. |
| Software Dependencies | No | Our presented hybrid voxel-octree method is implemented by C++. The refinement process is implemented by PyTorch with the ADAM optimizer. No specific version numbers for PyTorch or ADAM are mentioned. |
| Experiment Setup | Yes | For indoor scenes, the edge length of the voxel-octree leaf nodes is set to 0.05 m. The threshold Tmin is 0.02 m, and the threshold Tcur is set to 0.01. The learning rates for vertex positions, vertex colors, and spherical harmonics are set to 0.0001, 0.01, and 0.001, respectively. The weights for the loss terms are set to 50 for λlap, 1 for λnormal and 1 for λedge. We run 300 iterations to optimize each partial mesh. |