Reproduce “FOF: Learning Fourier Occupancy Field for Monocular Real-time Human Reconstruction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Qiao Feng, Yebin Liu, Yu-Kun Lai, Jingyu Yang, Kun Li
Researchers from Tianjin University, Tsinghua University, and Cardiff University introduce the Fourier Occupancy Field (FOF), a new 3D geometry representation for monocular human reconstruction. This representation enables the first real-time pipeline to generate high-fidelity explicit 3D meshes at over 30 frames per second from a single RGB image, achieving superior accuracy compared to prior methods.
Key insights
- Approximating 1D occupancy functions along rays with a few low-frequency Fourier series terms offers a compact, memory-efficient, and continuously reconstructible 3D representation.
- Transforming a 3D occupancy field into a 2D multi-channel image (FOF) allows for leveraging powerful 2D CNNs for fast and accurate learning from monocular images.
- The analytical and parallelizable conversions between FOF and mesh representations are computationally efficient, enabling real-time performance for both training label generation and inference.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.