SOV-CAD formulates CAD modeling-sequence reconstruction from images as a sequential decision-making problem. At each step, it uses the target’s orthographic projections, projections of the partially reconstructed model, and the active sketch as visual feedback. The framework applies offline reinforcement learning with a Decision Transformer and geometric-alignment rewards to select modeling actions. The authors report improvements over state-of-the-art CAD sequence reconstruction methods and strong data efficiency. Code is available on GitHub.
No heat snapshots are available in the last 24 hours.