Megadose AI progress, ranked and analyzed.

Roomer: Reflective Object-Grounded Model Editing and Repair for 3D Indoor Layout Synthesis

· HF Daily Papers ·
Roomer targets the leftover physical mistakes in generated rooms by tying each violation to the objects that caused it.

The framework represents a layout, reviews it for issues like collisions, blocked openings, out-of-bounds placement, and obstructed circulation, then proposes local edits. A deterministic solver checks candidate repairs before any change is accepted. The paper says edits are committed only after full-scene verification confirms the target problem is fixed without creating new hard violations or breaking protected constraints. The authors also introduce Roomer-CC for training and Roomer-Eval for testing validity and usability. HF Daily Papers' note

score 4

Categories: Research