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 Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Implicit-ARAP: Efficient Handle-Guided Neural Field Deformation via Local Patch Meshing
Authors: Daniele Baieri, Filippo Maggioli, Emanuele Rodolà, Simone Melzi, Zorah Laehner
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct a comprehensive evaluation showing that our method consistently outperforms baselines in deformation quality, robustness, and computational efficiency. |
| Researcher Affiliation | Academia | Daniele Baieri University of Milano-Bicocca EMAIL Filippo Maggioli Pegaso University EMAIL Emanuele Rodolà Sapienza University of Rome EMAIL Simone Melzi University of Milano-Bicocca EMAIL Zorah Lähner University of Bonn, Lamarr Institute EMAIL |
| Pseudocode | Yes | Algorithm 1 SDF zero level set sampling. 1: procedure REJECTPROJECTSAMPLING(fθ, N, τ, tmax) ... Algorithm 2 Implicit-ARAP training loop. 1: procedure TRAIN(T, λ1, λ2, n, m, ρ, k, fθ, dϕ, {(si, ti)}h i=1) ... |
| Open Source Code | Yes | Our codebase is available at this url. The code and part of our data (two TFD shapes) are provided in anonymized format as part of our supplementary materials. |
| Open Datasets | Yes | We employ two datasets in our evaluation: the first one (TFD) is obtained by designing a set of hand-crafted deformation experiments using mesh data from Thingi10k [59] and the Stanford 3D scanning repository. The second one is the De FAUST dataset introduced in [34]. |
| Dataset Splits | No | The paper mentions using "TFD" and "De FAUST" datasets for evaluation but does not specify any training, validation, or test splits. For example, "We employ two datasets in our evaluation: the first one (TFD) is obtained by designing a set of hand-crafted deformation experiments using mesh data from Thingi10k [59] and the Stanford 3D scanning repository. The second one is the De FAUST dataset introduced in [34]." |
| Hardware Specification | Yes | All of our experiments were run on a desktop computer with a 12GB NVIDIA RTX4070Ti GPU. |
| Software Dependencies | No | We implemented our algorithm in Python, relying on Py Torch [44] for neural network primitives, linear algebra and automatic differentiation. In addition, we used Polyscope [49] for visualization... |
| Experiment Setup | Yes | Where unspecified, all of our deformation experiments were run using the hyperparameters showed in Table 2. To train the neural SDFs, we used the architecture described in Section 6.2.1 with the hyperparameters listed in Table 3. The Adam optimizer runs for a total of 10000 steps and uses a starting learning rate of 10 4 and a scheduler which halves it at steps 1000, 2000, and 5000. |