Pith. sign in

REVIEW 7 cited by

Analyzing and Improving the Image Quality of StyleGAN

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 1912.04958 v2 pith:M5JGU57V submitted 2019-12-03 cs.CV cs.LGcs.NEeess.IVstat.ML

classification cs.CVcs.LGcs.NEeess.IVstat.ML
keywords imagequalitygeneratoradditionalarchitectureimprovingmodelmodeling
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

The style-based GAN architecture (StyleGAN) yields state-of-the-art results in data-driven unconditional generative image modeling. We expose and analyze several of its characteristic artifacts, and propose changes in both model architecture and training methods to address them. In particular, we redesign the generator normalization, revisit progressive growing, and regularize the generator to encourage good conditioning in the mapping from latent codes to images. In addition to improving image quality, this path length regularizer yields the additional benefit that the generator becomes significantly easier to invert. This makes it possible to reliably attribute a generated image to a particular network. We furthermore visualize how well the generator utilizes its output resolution, and identify a capacity problem, motivating us to train larger models for additional quality improvements. Overall, our improved model redefines the state of the art in unconditional image modeling, both in terms of existing distribution quality metrics as well as perceived image quality.

Discussion (0). Sign in to comment.

Forward citations

Cited by 7 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. crossMoDA Challenge: Evolution of Cross-Modality Domain Adaptation Techniques for Vestibular Schwannoma and Cochlea Segmentation from 2021 to 2023

    eess.IV 2025-06 conditional novelty 5.0 of 10

    The 2022 and 2023 crossMoDA challenge results show that training on heterogeneous multi-institutional data reduces segmentation outliers on homogeneous test sets, while cochlea Dice declined in 2023.

  2. Proto-LeakNet: Towards Signal-Leak Aware Attribution in Synthetic Human Face Imagery

    cs.CV 2025-11 reject novelty 4.0 of 10

    Proto-LeakNet reaches 98.13% closed-set Macro AUC but only about 57% AUROC for open-set separation, directly contradicting the abstract's claim of strong generalization.

  3. From Coarse to Fine: Learnable Discrete Wavelet Transforms for Efficient 3D Gaussian Splatting

    cs.CV 2025-06 conditional novelty 4.0 of 10

    AutoOpti3DGS uses learnable discrete wavelet transforms on input images to train 3DGS from coarse to fine, reducing peak Gaussian counts by about 18 to 23 percent with modest quality trade-offs.

  4. Masked Conditioning for Deep Generative Models

    cs.LG 2025-05 conditional novelty 4.0 of 10

    Masking conditions during training with varying sparsity schedules lets small VAEs and latent diffusion models generate engineering designs from partially specified inputs.

  5. Beyond Sliders: Mastering the Art of Diffusion-based Image Manipulation

    cs.CV 2025-09 reject novelty 3.0 of 10

    Beyond Sliders augments Concept Sliders with perceptual, adversarial, and an undefined triplet loss, claiming better real-world edits, but the evidence is weak and the derivation is not valid.

  6. Multimodal Generative AI with Autoregressive LLMs for Human Motion Understanding and Generation: A Way Forward

    cs.CV 2025-05 conditional novelty 3.0 of 10

    A survey paper reviews multimodal generative AI and autoregressive LLMs for text-driven human motion generation, with comparative tables of models, datasets, and metrics.

  7. Tackling fake images in cybersecurity -- Interpretation of a StyleGAN and lifting its black-box

    cs.CV 2025-07 reject novelty 2.0 of 10

    A StyleGAN trained on CelebA is pruned and its latent dimensions tweaked, showing many weights are redundant and some dimensions correlate with features, but results are anecdotal and lack external validation.

Pith tools