Reproduce “DarkVGGT: Seeing Through Darkness Using Thermal Geometry” 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
Minseong Kweon, Wenyuan Zhao, Nuo Chen, Lulin Liu, Huiwen Han, Zihao Zhu, Srinivas Shakkottai, Chao Tian and 1 more
DarkVGGT uses paired thermal images to improve low-light 3D reconstruction while retaining a pretrained RGB geometry model as its main estimator. CTFF splits thermal features into content-preserving and high-frequency pathways, while GSTR routes geometry-shared thermal information into RGB. Across three low-visibility benchmarks, mean depth AbsRel is 0.1373 versus 0.1630 for fine-tuned SEAR; daylight performance is close to VGGT on average, but not unchanged.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.