SPAD : Spatially Aware Multiview Diffusers
This work addresses the challenge of multi-view image generation for 3D content creation, offering full camera control and preventing issues like the multi-face Janus problem, though it is incremental in building upon existing diffusion models.
The authors tackled the problem of generating consistent multi-view images from text or single images by introducing SPAD, which extends a pretrained 2D diffusion model with cross-view interactions constrained by epipolar geometry and Plucker coordinates, achieving state-of-the-art results in novel view synthesis on datasets like Objaverse and Google Scanned Objects.
We present SPAD, a novel approach for creating consistent multi-view images from text prompts or single images. To enable multi-view generation, we repurpose a pretrained 2D diffusion model by extending its self-attention layers with cross-view interactions, and fine-tune it on a high quality subset of Objaverse. We find that a naive extension of the self-attention proposed in prior work (e.g. MVDream) leads to content copying between views. Therefore, we explicitly constrain the cross-view attention based on epipolar geometry. To further enhance 3D consistency, we utilize Plucker coordinates derived from camera rays and inject them as positional encoding. This enables SPAD to reason over spatial proximity in 3D well. In contrast to recent works that can only generate views at fixed azimuth and elevation, SPAD offers full camera control and achieves state-of-the-art results in novel view synthesis on unseen objects from the Objaverse and Google Scanned Objects datasets. Finally, we demonstrate that text-to-3D generation using SPAD prevents the multi-face Janus issue. See more details at our webpage: https://yashkant.github.io/spad