2606.06695

Total: 1

#1 S23DR 2026 Winning Solution [PDF] [Copy] [Kimi] [REL]

Authors: Jan Skvrna, Miroslav Purkrabek, Lukas Neumann

This text presents the winning solution to the S23DR 2026 challenge for structured 3D wireframe reconstruction from sparse SfM, fitted depth, and semantic segmentations. The method treats vertices as a conditional set and denoises 64 vertex tokens with a flow-matching DiT conditioned on Perceiver-style scene tokens. A global pass predicts the coarse structure, a hull-cropped second pass refines it, and a small multi-sample consensus step keeps the stochastic sampler well behaved. The final system ranked first on the private leaderboard, achievingHSS = 0.654.

Subject: Computer Vision and Pattern Recognition

Publish: 2026-06-04 20:19:59 UTC