MetaSSP: Enhancing Semi-supervised Implicit 3D Reconstruction through Meta-adaptive EMA and SDF-aware Pseudo-label Evaluation

TL;DR

MetaSSP addresses the challenge of limited labeled data in single-view 3D reconstruction by learning implicit SDF surfaces through semi-supervised learning. It combines gradient-guided parameter importance regularization for a stable meta-adaptive EMA teacher with an SDF-aware pseudo-label weighting that merges augmentation consistency and SDF variance. Beginning with a 10% labeled warm-up, the framework jointly optimizes labeled and unlabeled data and achieves state-of-the-art Pix3D results, reducing Chamfer Distance by about 20.6% and increasing IoU by about 24.1% over strong baselines. This approach enables scalable, high-fidelity implicit reconstruction from abundant unlabeled imagery for real-world applications.

Abstract

Implicit SDF-based methods for single-view 3D reconstruction achieve high-quality surfaces but require large labeled datasets, limiting their scalability. We propose MetaSSP, a novel semi-supervised framework that exploits abundant unlabeled images. Our approach introduces gradient-based parameter importance estimation to regularize adaptive EMA updates and an SDF-aware pseudo-label weighting mechanism combining augmentation consistency with SDF variance. Beginning with a 10% supervised warm-up, the unified pipeline jointly refines labeled and unlabeled data. On the Pix3D benchmark, our method reduces Chamfer Distance by approximately 20.61% and increases IoU by around 24.09% compared to existing semi-supervised baselines, setting a new state of the art.

Figures (3)

• Figure 1: MetaSSP Semi-Supervised 3D Reconstruction Pipeline: (a) Supervised Warm-up Stage; (b) Semi-Supervised Knowledge Distillation Stage; (c) Detailed Teacher–Student Meta-Adaptive EMA and Pseudo-Label Weighting Mechanism.
• Figure 2: Gradient-guided importance map computation.
• Figure 3: Results come from warm-up supervised, semi- supervised and ground truth.