尽管扩散模型（Diffusion Model）与流匹配（Flow Matching）已经把文本到图像生成（Text-to-Image, T2I）推向了更高的视觉质量与可控性，但他们通常在推理时需要数十步网络迭代，限制了其对于一些需要低延迟，Real-Time 的应用。 为了把推理步数降下来，现有路线通常 ...

本文详细介绍了Flow Matching这一新兴的生成建模方法，从数学理论基础出发，逐步构建完整的实现框架。与传统扩散模型通过逆向去噪过程生成数据不同，Flow Matching通过学习时间相关的速度场，建立从噪声分布到目标数据分布的直接映射路径。文章将理论推导与 ...

OpenAI 的 GPT-4o 在图像理解、生成和编辑任务上展现了顶级性能。流行的架构猜想是： Tokens → [Autoregressive 模型] → [Diffusion 模型] → 图像像素 该混合架构将自回归与扩散模型的优势结合。Salesforce Research、马里兰大学、弗吉尼亚理工、纽约大学、华盛顿大学的 ...

On Thursday, Stability AI announced Stable Diffusion 3, an open-weights next-generation image-synthesis model. It follows its predecessors by reportedly generating detailed, multi-subject images with ...

This course provides a comprehensive introduction to diffusion models and flow models for generative AI, covering both theoretical foundations and methodological advancements. The course is divided ...

A novel FlowViT-Diff framework that integrates a Vision Transformer (ViT) with an enhanced denoising diffusion probabilistic model (DDPM) for super-resolution reconstruction of high-resolution flow ...