Decoupling Global and Local Representations via Invertible Generative Flows

Authors: Xuezhe Ma, Xiang Kong, Shanghang Zhang, Eduard H Hovy

ICLR 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on standard image benchmarks demonstrate the effectiveness of our model in terms of density estimation, image generation and unsupervised representation learning.
Researcher Affiliation Academia 1University of Southern California 2Carnegie Mellon University 3University of California, Berkeley
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes The code for our model is available at https://github.com/Xuezhe Max/wolf.
Open Datasets Yes To evaluate our generative model, we conduct two groups of experiments on four benchmark datasets that are commonly used to evaluate deep generative models: CIFAR-10 (Krizhevsky & Hinton, 2009), 64 64 downsampled version Image Net (Oord et al., 2016), the bedroom category in LSUN (Yu et al., 2015) and the Celeb A-HQ dataset (Karras et al., 2018).
Dataset Splits No The paper mentions using standard benchmark datasets (CIFAR-10, Image Net, LSUN, Celeb A-HQ) but does not explicitly provide their specific training, validation, or test dataset splits (e.g., percentages or sample counts) within the main text or appendices.
Hardware Specification Yes Table 6 provides the number of parameters of different models on CIFAR-10, together with the corresponding training time over one epoch (measured on four Tesla V100 GPUs).
Software Dependencies No The paper mentions the use of the Adam optimizer and various architectures but does not provide specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes Table 5: Hyper-parameters in our experiments. Dataset batch size latent dim dz weight decay # updates of warmup CIFAR-10, 32 32 512 64 1e 6 50 Image Net, 64 64 256 128 5e 4 200 LSUN, 128 128 256 256 5e 4 200 Celeb A-HQ, 256 256 40 256 5e 4 200