Multiview Equivariance Improves 3D Correspondence Understanding with Minimal Feature Finetuning
This work addresses the limitation of vision foundation models in understanding 3D spatial relationships, which is crucial for applications in robotics and augmented reality, though it is incremental as it builds on existing models.
The paper tackled the problem of enhancing 3D awareness in ViT-based vision models by improving 3D equivariance, resulting in significant performance gains on tasks like pose estimation and tracking with minimal finetuning.
Vision foundation models, particularly the ViT family, have revolutionized image understanding by providing rich semantic features. However, despite their success in 2D comprehension, their abilities on grasping 3D spatial relationships are still unclear. In this work, we evaluate and enhance the 3D awareness of ViT-based models. We begin by systematically assessing their ability to learn 3D equivariant features, specifically examining the consistency of semantic embeddings across different viewpoints. Our findings indicate that improved 3D equivariance leads to better performance on various downstream tasks, including pose estimation, tracking, and semantic transfer. Building on this insight, we propose a simple yet effective finetuning strategy based on 3D correspondences, which significantly enhances the 3D correspondence understanding of existing vision models. Remarkably, finetuning on a single object for one iteration results in substantial gains. Our code is available at https://github.com/qq456cvb/3DCorrEnhance.