Resource-Constrained 3D Digital Modeling from Monocular Videos:From Rotation-Table Recordings to Interactive E-commerce Assets

Hemu He

Co-Presenters: Hemu He, Zining Jin, Leyang Luo

College: Hennings College of Science Mathematics and Technology

Major: BS.COMPUTER/SCI

Faculty Research Mentor: Kumar, Yulia  Li , Juan

Abstract:

Low-cost and scalable 3D construction has become increasingly important for consumer-facing applications such as online retail, virtual exhibition, and digital content platforms, where both visual fidelity and computational efficiency are critical. This work investigates 3D model construction from self-captured monocular videos, with a particular focus on rotation-table recordings of physical objects. Such a capture setting provides dense and structured multi-view observations while remaining accessible to non-expert users with minimal hardware requirements.We present a generic 3D construction pipeline that transforms rotation videos into interactive digital assets through feature extraction, multi-view representation learning, and geometry reconstruction. The pipeline supports the generation of rotatable 3D models that allow users to freely inspect objects from arbitrary viewpoints, while enabling semantic interaction at the component level. In particular, reconstructed models can be augmented with localized annotations that describe material properties, manufacturing processes, and design details, enhancing transparency and user engagement in online shopping scenarios.To ensure practical deployability, we analyze the construction process under explicit resource considerations, including computational cost, memory usage, inference latency, and storage footprint. Model quality is evaluated using a combination of geometric consistency, visual completeness, and view synthesis accuracy, allowing balanced assessment across different reconstruction strategies. Our results indicate that, beyond a moderate model capacity, improvements in capture quality and viewpoint coverage contribute more effectively to perceived model realism than increasing computational complexity.Overall, this work highlights the feasibility of building high-quality, interactive 3D digital representations from inexpensive video capture, and provides a flexible framework for integrating 3D construction into real-world e-commerce and visualization systems under realistic resource constraints.