This work presents a multimodal machine learning (MML) pipeline with zero-shot 3D completion for the digital preservation and authentication of Matryoshka nesting dolls (MND). A private collection is digitized as this novel multimodal dataset centered on turntable videos, augmented with single and group images and auxiliary physical and textual cues. A text modality is produced using Qwen3VL captions to enable video-text fusion and semantic motif analysis. A unimodal 2D baseline is established for fine-grained 8-way style recognition and a 3-way authenticity task, and is compared against multimodal configurations that incorporate learned text embeddings. To incorporate geometry as direct evidence, the pipeline integrates a silhouette-to-skeleton branch based on the Blum medial axis (BMA) and a convolutional autoencoder (CA) that reconstructs dense silhouettes from sparse skeletons, yielding a compact representation suitable for downstream 3D reasoning. The 3D pipeline is implemented along two complementary branches: zero-shot completion with a pretrained 3D prior (Hunyuan3D) and mesh-oriented skeletonization via a custom BMA procedure. Late fusion combines geometric and textual signals to improve decision confidence beyond appearance-only models. The framework supports authentication decisions with explicit geometric and semantic evidence and is transferable to other cultural artifacts. Potential applications include AR/VR, education, gaming, and assistive technologies. The code for this project is available upon request.
Authenticating cultural heritage artifacts such as Matryoshka Nesting Dolls (MNDs) is increasingly complicated by high-fidelity replicas that successfully mimic surface textures and palettes, leading traditional 2D computer vision models to exhibit dangerous overconfidence in false-positive classifications. To address this, we propose an auditable multimodal framework that transitions from appearance-only detection to a robust verification system based on the following three technical pillars: (1) a 2D visual stream utilizing a ConvNeXt-Tiny backbone for fine-grained style recognition; (2) a 3D geometric stream employing a custom 2D-to-3D reconstruction pipeline based on the Blum Medial Axis (BMA) and surfaces of revolution to capture axisymmetric structural fidelity; and (3) a semantic stream leveraging the Qwen3-VL vision-language model to generate human-interpretable evidence cards. To support this framework, we introduce a novel multimodal dataset comprising 168 unique physical MND sets and 27,387 labeled frames, archived for reproducibility. Our experimental results demonstrate that while 2D-only baselines achieve 77.9% authenticity accuracy, they suffer from a high Expected Calibration Error (ECE) of 0.121. The integrated multimodal framework achieves a superior authenticity accuracy of 96.7% and reduces the ECE to 0.041, representing a 66% improvement in calibration reliability. Crucially, the system shifts the mean confidence for incorrect replica classifications from a high-risk 0.82 to a safe 0.45.