Twin Auxilary Classifiers GAN

Authors: Mingming Gong, Yanwu Xu, Chunyuan Li, Kun Zhang, Kayhan Batmanghelich

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experimental results show that our TAC-GAN can successfully replicate the true data distributions on simulated data, and significantly improves the diversity of class-conditional image generation on real datasets. We first compare the distribution matching ability of AC-GAN, Projection c GAN, and our TAC-GAN on Mixture of Gaussian (Mo G) and MNIST [34] synthetic data. We evaluate the image generation performance of TAC-GAN on three image datatest including CIFAR100 [27], Image Net1000 [25] and VGGFace2 [28].
Researcher Affiliation Collaboration 1Department of Biomedical Informatics, University of Pittsburgh, {mig73,yanwuxu,kayhan}@pitt.edu 2Microsoft Research, Redmond, cl319@duke.edu 3Department of Philosophy, Carnegie Mellon University, kunz1@cmu.edu
Pseudocode No The paper does not include a figure, block, or section labeled 'Pseudocode' or 'Algorithm'.
Open Source Code Yes The code is available at https://github.com/batmanlab/twin_ac.
Open Datasets Yes We evaluate the image generation performance of TAC-GAN on three image datatest including CIFAR100 [27], Image Net1000 [25] and VGGFace2 [28]. We use the overlapping MNIST dataset to demonstrate the robustness of our TAC-GAN.
Dataset Splits Yes CIFAR100 [27] has 100 classes, each of which contains 500 training images and 100 testing images at the resolution of 32 32. We randomly sample from MNIST training set to construct two image groups: Group A contains 5,000 digit 1 and 5,000 digit 0 , while Group B contains 5,000 digit 2 and 5,000 digit 0 ,to simulate overlapping distributions, where digit 0 appears in both groups.
Hardware Specification Yes We gratefully acknowledge the support of NVIDIA Corporation with the donation of the Titan X Pascal GPU used for this research. We were also grateful for the computational resources provided by Pittsburgh Super Computing grant number TG-ASC170024.
Software Dependencies No We implemented TAC-GAN in Pytorch.
Experiment Setup No The detailed experiment setups are shown in the SM.