MeshSDF: Differentiable Iso-Surface Extraction

Authors: Edoardo Remelli, Artem Lukoianov, Stephan Richter, Benoit Guillard, Timur Bagautdinov, Pierre Baque, Pascal Fua

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We use two different applications to validate our theoretical insight: Single-View Reconstruction via Differentiable Rendering and Physically-Driven Shape Optimization. In both cases our differentiable parameterization gives us an edge over state-of-the-art algorithms.
Researcher Affiliation Collaboration Edoardo Remelli 1 Artem Lukoianov 1,2 Stephan R. Richter 3 Benoît Guillard 1 Timur Bagautdinov 2 Pierre Baque 2 Pascal Fua 1 1CVLab, EPFL, {name.surname}@epfl.ch 2Neural Concept SA, {name.surname}@neuralconcept.com 3Intel Labs, {name.surname}@intel.com
Pseudocode Yes Algorithm 1: Mesh SDF Forward
Open Source Code No The paper cites PyTorch3D [38] with a link, but this is a third-party library they used. There is no explicit statement or link from the authors stating they are releasing their Mesh SDF code.
Open Datasets Yes We demonstrate that our method is straightforward to apply to this task and validate our approach on two standard datasets, namely Shape Net [6] and Pix3D [45].
Dataset Splits Yes We used standard train/test splits along with the renderings provided in [56] for all the methods we tested.
Hardware Specification No No specific GPU, CPU, or other hardware details are mentioned for running the experiments.
Software Dependencies No The paper mentions 'Open Foam [18]' and 'PyTorch3D [38]' and 'Adam [23]', but does not provide specific version numbers for any of these software components as required.
Experiment Setup Yes In practice, we run 400 gradient descent iterations using Adam [23] and keep the z with the smallest Ltask as our final code vector.