按年被提到
20232023 · 2 次2026
视觉、语音与图像生成›图像与视频生成·arXiv:2303.05511 · cs.CV

Scaling up GANs for Text-to-Image Synthesis

Minguk Kang, Jun-Yan Zhu, Richard Zhang, Jaesik Park, Eli Shechtman, Sylvain Paris, Taesung Park · 2023
开始阅读 ↗PDFalphaXiv 讨论
被提及 · 2 人首次 2023-08-23 · 最近 2023-12-14
2不同的人
2活跃月份
2讨论串
425被引(OpenAlex)
arXiv 摘要(候选层不写导读)

The recent success of text-to-image synthesis has taken the world by storm and captured the general public's imagination. From a technical standpoint, it also marked a drastic change in the favored architecture to design generative image models. GANs used to be the de facto choice, with techniques like StyleGAN. With DALL-E 2, auto-regressive and diffusion models became the new standard for large-scale generative models overnight. This rapid shift raises a fundamental question: can we scale up GANs to benefit from large datasets like LAION? We find that naÏvely increasing the capacity of the StyleGAN architecture quickly becomes unstable. We introduce GigaGAN, a new GAN architecture that far exceeds this limit, demonstrating GANs as a viable option for text-to-image synthesis. GigaGAN offers three major advantages. First, it is orders of magnitude faster at inference time, taking only 0.13 seconds to synthesize a 512px image. Second, it can synthesize high-resolution images, for example, 16-megapixel pixels in 3.66 seconds. Finally, GigaGAN supports various latent space editing applications such as latent interpolation, style mixing, and vector arithmetic operations.

候选层:至少 2 个不同的人、在 2 个不同月份提到过,还没到主书架的 3 人 3 个月。没有做立场标注,所以只写“被提及”。

HN 书评

推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及
评论gwern · 2023-08

“If you're wondering how it's so fast and cheap and they can generate variants so easily, it's because they're using GANs (see the footer). GANs are way faster than diffusion models because they generate the image in a single forward pass and their true latent space encoding makes editing a breeze. (And if you're wondering how it can look so good when 'ev…”

AI Real-Time Human Full-Body Photo Generator原评论 ↗
评论godelski · 2023-12

“Sorry, I left too much knowledge for inference. I presumed it was safe to let users infer that a 2018 (StyleGAN) or 2019 (StyleGAN2) model was not going to compete with performance of a 2023 model, regardless of architecture. There has in fact been improvements in GANs in the last 5 years. Text conditioning is not unique to diffusion and is rather a subn…”

Stable Diffusion Turbo on a Raspberry Pi Zero 2 generates an image in 29 minutes原评论 ↗