Researchers have developed a novel Video Diffusion Model called Rays as Pixels, which learns a joint distribution of videos and camera trajectories. This model is the first to predict camera poses and generate videos from novel viewpoints within a single framework. By representing cameras as ray pixels (raxels) in the same latent space as video frames, the model can perform tasks such as predicting camera trajectories, generating video from input images along a defined path, and jointly synthesizing video and trajectory. AI
IMPACT Introduces a unified approach for video and camera trajectory learning, potentially improving scene understanding and novel view synthesis.
RANK_REASON This is a research paper describing a novel AI model and its capabilities. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
