Manifold-Valued Image Generation with Wasserstein Generative Adversarial Nets

Authors: Zhiwu Huang, Jiqing Wu, Luc Van Gool3886-3893

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental On the three datasets, we experimentally demonstrate the proposed manifold-aware WGAN model can generate more plausible manifold-valued images than its competitors.
Researcher Affiliation Academia Computer Vision Lab, ETH Zurich, Switzerland VISICS, KU Leuven, Belgium
Pseudocode Yes Algorithm 1 Manifold-aware Wasserstein GAN (manifold WGAN), our proposed algorithm.
Open Source Code No The paper mentions 'The official code is available at https://github.com/igul222/improved_wgan_training' which refers to an existing WGAN implementation and not the code for the authors' proposed method.
Open Datasets Yes For the studied manifold-valued image generation problem, we suggest three benchmark evaluations that use the HSV and CB images of the well-known CIFAR-10 (Krizhevsky and Hinton 2009), Image Net (Oord, Kalchbrenner, and Kavukcuoglu 2016), and the popular UCL DT image dataset (Cook et al. 2006).
Dataset Splits Yes We use the 64 64 version of Image Net, which contains 1,281,149 training images and 49,999 images for testing.
Hardware Specification No The paper mentions 'We would like to thank Nvidia for donating the GPUs used in this work.' but does not specify any particular GPU models or other hardware details.
Software Dependencies No The paper does not provide specific software dependencies with version numbers, such as programming language versions or library versions.
Experiment Setup Yes We finally optimize the network using Adam with learning rate 0.0002, decayed linearly to 0 over 100K generator iterations, and batch size 64.