New Wavelet Transformer Achieves State-of-the-Art Image Generation

Researchers have developed WaiT, a new wavelet-aware image Transformer that improves image generation quality by addressing the uniform treatment of spatial frequencies in standard models. WaiT decomposes generation into coarse and fine bands, allowing high-frequency details to be refined only after coarse structures have emerged. This approach leads to a new state-of-the-art FID of 1.3 on ImageNet 512x512 and a competitive FVD of 0.84 on Kinetics-600 for video generation, while also reducing sampling compute by up to 50%. AI

IMPACT This research introduces a novel approach to image generation that could lead to more efficient and higher-quality AI-powered visual content creation.

RANK_REASON The cluster describes a new academic paper detailing a novel model architecture and benchmark results. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv stat.ML →

• ImageNet
• Kinetics-600
• Krunoslav Lehman Pavasovic
• OpenImages
• Transformer++

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →