MGAN: Training Generative Adversarial Nets with Multiple Generators

Authors: Quan Hoang, Tu Dinh Nguyen, Trung Le, Dinh Phung

ICLR 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive experiments on synthetic 2D data and natural image databases (CIFAR-10, STL-10 and Image Net) to demonstrate the superior performance of our MGAN in achieving state-of-the-art Inception scores over latest baselines, generating diverse and appealing recognizable objects at different resolutions, and specializing in capturing different types of objects by the generators.
Researcher Affiliation Academia Quan Hoang University of Massachusetts-Amherst Amherst, MA, USA qhoang@umass.edu Tu Dinh Nguyen, Trung Le, Dinh Phung PRa DA Centre, Deakin University Geelong, Australia {tu.nguyen,trung.l,dinh.phung} @deakin.edu.au
Pseudocode Yes We refer to Appendix A for the pseudo-code and algorithms for parameter learning for our proposed MGAN.
Open Source Code Yes We use Tensor Flow (Abadi et al., 2016) to implement our model, and the source code is available at: https://github.com/qhoangdl/MGAN.
Open Datasets Yes We use 3 widely-adopted datasets: CIFAR-10 (Krizhevsky & Hinton, 2009), STL-10 (Coates et al., 2011) and Image Net (Russakovsky et al., 2015).
Dataset Splits Yes We use 3 widely-adopted datasets: CIFAR-10 (Krizhevsky & Hinton, 2009), STL-10 (Coates et al., 2011) and Image Net (Russakovsky et al., 2015). CIFAR-10 contains 50,000 32 32 training images of 10 classes: airplane, automobile, bird, cat, deer, dog, frog, horse, ship, and truck.
Hardware Specification No The paper does not explicitly describe the hardware used to run its experiments with specific details such as GPU models, CPU models, or memory.
Software Dependencies No The paper mentions using 'Tensor Flow (Abadi et al., 2016)' but does not specify its version or any other software dependencies with version numbers.
Experiment Setup Yes For all experiments, we use: (i) shared parameters among generators in all layers except for the input layer; (ii) shared parameters between discriminator and classifier in all layers except for the output layer; (iii) Adam optimizer (Kingma & Ba, 2014) with learning rate of 0.0002 and the first-order momentum of 0.5; (iv) minibatch size of 64 samples for training discriminators; (v) Re LU activations (Nair & Hinton, 2010) for generators; (vi) Leaky Re LU (Maas et al., 2013) with slope of 0.2 for discriminator and classifier; and (vii) weights randomly initialized from Gaussian distribution N(0, 0.02I) and zero biases.