OOGAN: Disentangling GAN with One-Hot Sampling and Orthogonal Regularization
Authors: Bingchen Liu, Yizhe Zhu, Zuohui Fu, Gerard de Melo, Ahmed Elgammal4836-4843
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct quantitative and qualitative experiments to demonstrate the advantages of our method on several datasets. |
| Researcher Affiliation | Academia | Department of Computer Science Rutgers University {bingchen.liu, yizhe.zhu, zuohui.fu, gerard.demelo}@rutgers.edu, elgammal@cs.rutgers.edu |
| Pseudocode | No | The paper includes diagrams of model structures but does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | All the code to reproduce our experiments is available on Git Hub, and training configurations can be found there. |
| Open Datasets | Yes | quantitative experiments on the d Sprites datasets (Matthey et al. 2017) following the metric proposed by Kim and Mnih (2018). After that, we show the superiority of OOGAN in generating high-quality images while maintaining competitive disentanglement compared to VAE-based models on Celeb A (Liu et al. 2015) and 3D-chair (Aubry etal. 2014) data. |
| Dataset Splits | No | The paper uses standard datasets and refers to setups from prior work (Kim and Mnih 2018; Jeon, Lee, and Kim 2019), but it does not explicitly state the specific training, validation, and test split percentages or sample counts within its own text. |
| Hardware Specification | Yes | We perform all the experiments on one NVIDIA RTX 2080Ti GPU |
| Software Dependencies | No | The paper states that code is available on GitHub with training configurations but does not explicitly list specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | On the 3D Chairs data, we use 64x64 RGB images with batch size 64 for all training runs. and where varying the hyperparameters of λ (1 to 5) and γ (0.2 to 2) in our loss function always yields consistent performance. |