Rank3DGAN: Semantic Mesh Generation Using Relative Attributes
Authors: Yassir Saquil, Qun-Ce Xu, Yong-Liang Yang, Peter Hall5586-5594
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical results In this section, we describe our experimental settings, show quantitative and qualitative results, and then demonstrate applications in mesh generation, mesh editing, and mesh attribute transfer. We used the metric Fr echet Inception Distance (FID) (Heusel et al. 2017) for evaluating the quality of generated charts. We also compared Rank3DGAN against voxel 3D GAN (Wu et al. 2016) and Atlas Net (Groueix et al. 2018) using the volumetric Io U, Chamfer distance and normal consistency metrics on human, face, bird datasets. |
| Researcher Affiliation | Academia | Yassir Saquil,* Qun-Ce Xu, Yong-Liang Yang, Peter Hall University of Bath, Bath, UK {ys999, qx289, yy753, maspmh}@bath.ac.uk |
| Pseudocode | No | The paper does not contain pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | No | The paper does not provide any explicit statement or link for open-source code of the methodology. |
| Open Datasets | Yes | We relied on four datasets for our experiments: 3D human shapes from MPII Human Shape (Pishchulin et al. 2017), 3D bird meshes reconstructed from CUB image dataset (Kanazawa et al. 2018), and 3D faces from Basel Face Model 2009 (Paysan et al. 2009) and 2017 (Gerig et al. 2018). |
| Dataset Splits | No | The paper mentions training on datasets but does not provide specific training/validation/test splits, percentages, or sample counts needed to reproduce the data partitioning. |
| Hardware Specification | No | The paper does not provide specific hardware details (such as exact GPU/CPU models or processor types) used for running its experiments. |
| Software Dependencies | No | The paper mentions implementing Rank3DGAN on top of multi-chart 3D GAN and using FCN, but does not provide specific version numbers for these or any other software dependencies. |
| Experiment Setup | Yes | For the training, we set the hyperparameters λ = 10, ν = 1, and trained the networks for 300 epochs on all datasets. Similarly to (Hamu et al. 2018), we activated the landmark consistency layer after 50 training epochs for human and bird datasets. |