Neural Unsigned Distance Fields for Implicit Function Learning
Authors: Julian Chibane, Mohamad Aymen mir, Gerard Pons-Moll
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on Shape Net [13] show that NDF, while simple, is the state-of-the art, and allows to reconstruct shapes with inner structures, such as the chairs inside a bus. |
| Researcher Affiliation | Academia | Julian Chibane Aymen Mir Gerard Pons-Moll Max Planck Institute for Informatics, Saarland Informatics Campus, Germany {jchibane,amir,gpons}@mpi-inf.mpg.de |
| Pseudocode | Yes | Algorithm 1: NDF: Dense PCs |
| Open Source Code | Yes | To encourage further research in this new direction, we make code and model publicly available at https://virtualhumans.mpi-inf.mpg.de/ndf/. |
| Open Datasets | Yes | Experiments on Shape Net [13] show that NDF, while simple, is the state-of-the art, and allows to reconstruct shapes with inner structures, such as the chairs inside a bus. [...] Angel X Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio Savarese, Manolis Savva, Shuran Song, Hao Su, Jianxiong Xiao, Li Yi, and Fisher Yu. Shapenet: An information-rich 3d model repository. ar Xiv preprint ar Xiv:1512.03012, 2015. |
| Dataset Splits | Yes | We use a 80/20 random train and test split. |
| Hardware Specification | Yes | Each projection takes around 3.7 seconds for a 1 million points on a Tesla V100. |
| Software Dependencies | No | The paper does not list specific software dependencies (e.g., libraries, frameworks) with their version numbers. |
| Experiment Setup | Yes | clamping the maximal regressed distance to a value δ > 0 concentrates the model capacity to represent the vicinity of the surface accurately. Larger δ values increase the convergence of our visualization Algorithms 1 and 2. We find a good trade-offwith δ = 10cm. [...] To address 1), we found that projecting a point with Eq. 2 multiple times (we use 5 in our experiments) with unit normalized gradient yields accurate surface predictions. |