ViDAR: Video Diffusion-Aware 4D Reconstruction From Monocular Inputs
This addresses the challenging problem of generating photorealistic views of moving subjects from arbitrary viewpoints for computer vision researchers, representing a strong incremental advance over existing methods.
The paper tackles dynamic novel view synthesis from monocular video by introducing ViDAR, a 4D reconstruction framework that uses personalized diffusion models to generate pseudo multi-view supervision for training Gaussian splatting representations, achieving state-of-the-art performance on the DyCheck benchmark with improved visual quality and geometric consistency.
Dynamic Novel View Synthesis aims to generate photorealistic views of moving subjects from arbitrary viewpoints. This task is particularly challenging when relying on monocular video, where disentangling structure from motion is ill-posed and supervision is scarce. We introduce Video Diffusion-Aware Reconstruction (ViDAR), a novel 4D reconstruction framework that leverages personalised diffusion models to synthesise a pseudo multi-view supervision signal for training a Gaussian splatting representation. By conditioning on scene-specific features, ViDAR recovers fine-grained appearance details while mitigating artefacts introduced by monocular ambiguity. To address the spatio-temporal inconsistency of diffusion-based supervision, we propose a diffusion-aware loss function and a camera pose optimisation strategy that aligns synthetic views with the underlying scene geometry. Experiments on DyCheck, a challenging benchmark with extreme viewpoint variation, show that ViDAR outperforms all state-of-the-art baselines in visual quality and geometric consistency. We further highlight ViDAR's strong improvement over baselines on dynamic regions and provide a new benchmark to compare performance in reconstructing motion-rich parts of the scene. Project page: https://vidar-4d.github.io