Diffusion Models For Multi-Modal Generative Modeling
This work addresses the need for more generalizable generative models in AI by enabling multi-modal training, though it is incremental as it builds on existing diffusion model paradigms.
The paper tackles the problem of extending diffusion models from single-modal to multi-modal generative modeling by proposing a unified framework that constructs a common diffusion space and uses a shared backbone denoising network with modality-specific decoders. The result is effective performance in various multi-modal generation settings, such as image transition and joint image-label modeling, as demonstrated by extensive experiments on ImageNet.
Diffusion-based generative modeling has been achieving state-of-the-art results on various generation tasks. Most diffusion models, however, are limited to a single-generation modeling. Can we generalize diffusion models with the ability of multi-modal generative training for more generalizable modeling? In this paper, we propose a principled way to define a diffusion model by constructing a unified multi-modal diffusion model in a common diffusion space. We define the forward diffusion process to be driven by an information aggregation from multiple types of task-data, e.g., images for a generation task and labels for a classification task. In the reverse process, we enforce information sharing by parameterizing a shared backbone denoising network with additional modality-specific decoder heads. Such a structure can simultaneously learn to generate different types of multi-modal data with a multi-task loss, which is derived from a new multi-modal variational lower bound that generalizes the standard diffusion model. We propose several multimodal generation settings to verify our framework, including image transition, masked-image training, joint image-label and joint image-representation generative modeling. Extensive experimental results on ImageNet indicate the effectiveness of our framework for various multi-modal generative modeling, which we believe is an important research direction worthy of more future explorations.