近日，Adobe研究院与斯坦福大学等机构的研究者联手发布了一种新的3D生成方法——“DMV3D”。这一方法采用基于Transformer的3D大型重建模型，对多视图扩散进行去噪，从而实现高效、高质量的3D图像生成。
据悉，DMV3D能够直接根据模型文字或单张图片的输入，生成3D NeRF。在单个A100 GPU上，仅需30秒，DMV3D就能生成各种高保真3D图像。这一成果对于3D建模、虚拟现实、增强现实等领域具有重要的应用价值。
此次研究的成功，展示了人工智能在3D生成领域的强大潜力。未来，随着技术的不断进步，3D生成技术将在更多场景中得到广泛应用，为人们的生活带来更多便利。
英文翻译：
News Title: Adobe and Stanford University Collaborate on Efficient 3D Generation Method
Keywords: 3D generation, DMV3D, Adobe Research Institute, Stanford University
News Content:
Recently, researchers from Adobe Research Institute and Stanford University have jointly developed a new 3D generation method called “DMV3D”. This method uses a 3D large-scale reconstruction model based on transformers to denoise multi-view diffusion, achieving efficient and high-quality 3D image generation.
It is reported that DMV3D can directly generate 3D NeRF based on model text or single-image input. On a single A100 GPU, it only takes 30 seconds to generate various high-fidelity 3D images. This achievement has important applications in fields such as 3D modeling, virtual reality, and augmented reality.
The success of this research demonstrates the strong potential of artificial intelligence in the field of 3D generation. In the future, with the continuous advancement of technology, 3D generation technology will be widely applied in more scenarios, bringing more convenience to people’s lives.
【来源】https://new.qq.com/rain/a/20231127A04PHM00
Views: 1