MM-3DScene: 3D Scene Understanding by Customizing Masked Modeling with Informative-Preserved Reconstruction and Self-Distilled Consistency
This work addresses the problem of 3D scene understanding for computer vision applications, offering an incremental advancement over existing masked modeling techniques by adapting them to handle data sparsity and complexity in 3D scenes.
The paper tackles the challenge of applying Masked Modeling to large-scale 3D scenes by proposing a novel informative-preserved reconstruction method that preserves representative structured points and integrates progressive reconstruction with self-distilled consistency, resulting in improvements such as +6.1 mAP@0.5 on object detection and +2.2% mIoU on semantic segmentation.
Masked Modeling (MM) has demonstrated widespread success in various vision challenges, by reconstructing masked visual patches. Yet, applying MM for large-scale 3D scenes remains an open problem due to the data sparsity and scene complexity. The conventional random masking paradigm used in 2D images often causes a high risk of ambiguity when recovering the masked region of 3D scenes. To this end, we propose a novel informative-preserved reconstruction, which explores local statistics to discover and preserve the representative structured points, effectively enhancing the pretext masking task for 3D scene understanding. Integrated with a progressive reconstruction manner, our method can concentrate on modeling regional geometry and enjoy less ambiguity for masked reconstruction. Besides, such scenes with progressive masking ratios can also serve to self-distill their intrinsic spatial consistency, requiring to learn the consistent representations from unmasked areas. By elegantly combining informative-preserved reconstruction on masked areas and consistency self-distillation from unmasked areas, a unified framework called MM-3DScene is yielded. We conduct comprehensive experiments on a host of downstream tasks. The consistent improvement (e.g., +6.1 mAP@0.5 on object detection and +2.2% mIoU on semantic segmentation) demonstrates the superiority of our approach.