VisCo Grids: Surface Reconstruction with Viscosity and Coarea Grids

Authors: Albert Pumarola, Artsiom Sanakoyeu, Lior Yariv, Ali Thabet, Yaron Lipman

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section we extensively evaluate Vis Co grids. First, we evaluate on two standard surface reconstruction benchmarks [46, 22] (Sec. 4.1) against a large variety of state-of-the-art methods: Poisson Surface Reconstruction [24], DGP [46], IGR [18], SIREN [43], FFN [45], NSP [47], PHASE [27], GD [14], BPA [8], SPSR [25], RIMLS [34], SALD [5], IGR [19], Occ Net [30], Deep SDF [36], LIG [23], Points2Surf [17], DSE [38], IMLSNet [28] and Parse Net [42]. We then perform an ablation study (Sec. 4.2), and conduct a detail examination of the main components of the model, namely the viscosity and coarea losses. Finally, we discuss the model s ability to reconstruct sparse point clouds (Sec. 4.3) using scans from Stanford 3D Scanning Repository.
Researcher Affiliation Collaboration 1Meta AI, 2Weizmann Institute of Science
Pseudocode No The paper describes methods and formulas but does not include any explicit pseudocode or algorithm blocks.
Open Source Code No No explicit statement about open-sourcing code or providing a repository link for the methodology.
Open Datasets Yes We next evaluated our model on two benchmarks: Surface Reconstruction Benchmark [46] and Surface Reconstruction from Real-Scans [22].
Dataset Splits No The paper mentions using 'standard' benchmarks and that 'we use same hyper-paramenters for all meshes of all benchmarks', but it does not specify exact train/validation/test splits (e.g., percentages or counts) or provide citations to predefined splits.
Hardware Specification No No specific hardware details (e.g., CPU, GPU models, or memory) used for experiments are mentioned in the paper.
Software Dependencies No No specific software dependencies with version numbers are mentioned in the paper.
Experiment Setup No The paper mentions key hyperparameters of the loss function (λp, λn, λv, λc) and discusses varying ϵ and λc in ablation studies, but it does not provide specific values for all hyperparameters, training configurations, or system-level settings required for full experimental setup reproduction (e.g., learning rate, optimizer, batch size, epochs, etc.).