You Only Need Adversarial Supervision for Semantic Image Synthesis

Authors: Edgar Schönfeld, Vadim Sushko, Dan Zhang, Juergen Gall, Bernt Schiele, Anna Khoreva

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

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct experiments on three challenging datasets: ADE20K (Zhou et al., 2017), COCO-stuff (Caesar et al., 2018) and Cityscapes (Cordts et al., 2016). Following prior work (Isola et al., 2017; Wang et al., 2018; Park et al., 2019; Liu et al., 2019), we evaluate models quantitatively on the validation set using the Fr echet Inception Distance (FID) (Heusel et al., 2017) and mean Intersection-over-Union (m Io U).
Researcher Affiliation Collaboration Edgar Sch onfeld Bosch Center for Artificial Intelligence Vadim Sushko * Bosch Center for Artificial Intelligence Dan Zhang Bosch Center for Artificial Intelligence J urgen Gall University of Bonnartificialintelligence Bernt Schiele Max Planck Institute for Informaticsartificialintelligen Anna Khoreva Bosch Center for Artificial Intelligence
Pseudocode No The paper provides detailed architectural tables for the discriminator and generator, but it does not include any blocks or sections explicitly labeled 'Pseudocode' or 'Algorithm'.
Open Source Code Yes Our code and pretrained models are available at https://github.com/boschresearch/OASIS.
Open Datasets Yes We conduct experiments on three challenging datasets: ADE20K (Zhou et al., 2017), COCO-stuff (Caesar et al., 2018) and Cityscapes (Cordts et al., 2016).
Dataset Splits Yes Following prior work (Isola et al., 2017; Wang et al., 2018; Park et al., 2019; Liu et al., 2019), we evaluate models quantitatively on the validation set using the Fr echet Inception Distance (FID) (Heusel et al., 2017) and mean Intersection-over-Union (m Io U).
Hardware Specification Yes All the experiments were run on 4 Tesla V100 GPUs, with a batch size of 20 for Cityscapes, and 32 for ADE20k and COCO-Stuff.
Software Dependencies No The paper mentions using the Adam optimizer but does not provide specific version numbers for software dependencies such as programming languages, deep learning frameworks, or other libraries.
Experiment Setup Yes The Adam (Kingma & Ba, 2015) optimizer was used with momentums β = (0, 0.999) and constant learning rates (0.0001, 0.0004) for G and D. We did not apply the GAN feature matching loss, and used the VGG perceptual loss only for ablations with λVGG = 10. The coefficient for Label Mix λLM was set to 5 for ADE20k and Cityscapes, and to 10 for COCO-Stuff. All our models use an exponential moving average (EMA) of the generator weights with 0.9999 decay (Brock et al., 2019). All the experiments were run on 4 Tesla V100 GPUs, with a batch size of 20 for Cityscapes, and 32 for ADE20k and COCO-Stuff. The training epochs are 200 on ADE20K and Cityscapes, and 100 for the larger COCO-Stuff dataset.