Diffusion Mamba

Diffusion Mamba (DiM) is a novel architecture designed to improve the efficiency and scalability of diffusion models, particularly for high-resolution image and video generation, as well as 3D shape modeling. Current research focuses on adapting the Mamba state-space model to overcome the computational limitations of traditional transformer-based diffusion models, employing techniques like selective scanning and bidirectional processing to achieve linear complexity. This approach offers significant advantages in speed and memory usage, enabling the generation of high-fidelity outputs in applications ranging from medical image conversion to realistic motion style transfer, thereby advancing the capabilities of generative models across diverse fields.