Meissonic：重振掩码生成式 Transformer 以实现高效的高分辨率文本到图像合成
Meissonic: Revitalizing Masked Generative Transformers for Efficient High-Resolution Text-to-Image Synthesis
AI总结：
本文提出 Meissonic，通过架构创新、位置编码、采样优化、偏好微条件与特征压缩，使非自回归 MIM 文生图在效率和 $1024 \ imes 1024$ 质量上匹敌甚至超越 SDXL。
摘要：
我们提出 Meissonic，它将非自回归掩码图像建模（MIM）文本到图像提升到可与 SDXL 等最先进扩散模型相媲美的水平。通过整合一整套架构创新、先进的位置编码策略和优化的采样条件，Meissonic 显著提升了 MIM 的性能与效率。此外，我们利用高质量训练数据，融合由人类偏好分数提供依据的微条件，并采用特征压缩层，以进一步增强图像保真度和分辨率。我们的模型在生成高质量、高分辨率图像方面不仅匹敌，而且常常超过 SDXL 等现有模型的性能。大量实验验证了 Meissonic 的能力，表明其有潜力成为文本到图像合成的新标准。我们发布了一个能够生成 $1024 \ imes 1024$ 分辨率图像的模型检查点。
英文摘要：
We present Meissonic, which elevates non-autoregressive masked image modeling (MIM) text-to-image to a level comparable with state-of-the-art diffusion models like SDXL. By incorporating a comprehensive suite of architectural innovations, advanced positional encoding strategies, and optimized sampling conditions, Meissonic substantially improves MIM's performance and efficiency. Additionally, we leverage high-quality training data, integrate micro-conditions informed by human preference scores, and employ feature compression layers to further enhance image fidelity and resolution. Our model not only matches but often exceeds the performance of existing models like SDXL in generating high-quality, high-resolution images. Extensive experiments validate Meissonic's capabilities, demonstrating its potential as a new standard in text-to-image synthesis. We release a model checkpoint capable of producing $1024 \times 1024$ resolution images.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
