Reproduce “HEAT: Holistic Edge Attention Transformer for Structured 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
Jiacheng Chen, Yiming Qian, Yasutaka Furukawa
HEAT reconstructs planar graphs from images by treating each candidate connection between detected corners as a Transformer node, allowing the model to reason across edges. Shared decoder branches support learning graph structure. On 256×256 satellite crops, HEAT achieves region F1 of 70.6, versus 63.4 for HAWP. Evidence is limited to planar-graph tasks, with missed corners and rare structures remaining failure cases.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.