diffusion model
PulseAugur coverage of diffusion model — every cluster mentioning diffusion model across labs, papers, and developer communities, ranked by signal.
6 天有情绪数据
-
UniSER基础模型统一图像柔光效果去除
研究人员开发了UniSER，这是一种新颖的基础模型，旨在解决数字图像中的各种柔光视觉退化问题，例如镜头眩光、薄雾、阴影和反射。与以往单独处理这些问题的专用模型不同，UniSER提供了一个统一的框架。这是通过一个包含380万图像对的海量数据集和一个经过微调的Diffusion Transformer实现的，能够实现超越现有专用和通用方法的、稳健且高保真的图像恢复。
-
MetaSR框架使用Diffusion Transformer进行生成式超分辨率中的自适应元数据
研究人员开发了MetaSR，一个用于生成式超分辨率的新型框架，该框架能够自适应地选择和注入相关元数据以增强图像和视频质量。这种基于Diffusion Transformer的方法旨在处理各种内容和降级场景，性能优于现有方法。MetaSR在PSNR方面取得了显著改进，同时在资源受限的情况下将传输比特率降低了多达50%。
-
Audio-Omni framework unifies audio generation, editing, and understanding
Researchers have introduced Audio-Omni, a novel framework designed to unify audio understanding, generation, and editing across diverse domains like speech, music, and general sounds. This system integrates a frozen Mul…
-
新的REDEdit框架通过扩散Transformer实现无掩码局部图像编辑
研究人员开发了REDEdit，一个新颖的适配器框架，旨在提高大型扩散Transformer（DiTs）中局部图像编辑的精度。该系统可以改造现有的DiTs，而无需改变其核心权重，使其能够在指定区域内精确执行编辑。REDEdit通过注入结构化条件流来分离编辑指令和空间位置，利用学习到的SpatialGate进行选择性信号路由，并使用区域感知损失（Region-Aware Loss）将训练集中在修改后的像素上。这种方法在部署时消除了对用户提…
-
Researchers develop adaptive diffusion for AI models to resist image corruptions
Researchers have developed a new framework for adapting AI models to handle image corruptions during testing, without needing to retrain the original model. This method uses a diffusion model to remove artifacts caused …
-
AI research advances 3D asset generation and anomaly detection for autonomous driving
Researchers have developed a novel approach called GenAssets for generating high-quality 3D assets from in-the-wild LiDAR and camera data, crucial for autonomous driving simulations. This method utilizes a "reconstruct-…
-
Generative AI creates synthetic malware to boost cybersecurity defenses
Researchers have developed a novel system to generate synthetic malware samples using generative AI, addressing the challenge of scarce and imbalanced datasets in cybersecurity. By treating malware binaries as mnemonic …