Presentation
BoxCtrl: 3D-Aware Visual Prompting for Geometric Image Editing
SessionImage Editing
DescriptionBoxCtrl introduces a 3D-aware visual prompting framework for precise image editing. Using RGB 3D bounding boxes to decouple geometry from appearance enables accurate translation, rotation, and scaling. A two-stage training of first SFT and then Reinforcement Learning achieves superior precision and photorealism in geometric editing across diverse real-world data.

Event Type
Technical Paper
TimeThursday, 23 July 202610:45am - 10:55am PDT
LocationRoom 408 B
Digital Library
PDF
Session Time & Location
Sunday, 19 July 20266:00pm - 8:45pm PDTHall K
Thursday, 23 July 202610:45am - 12:15pm PDTRoom 408 B
Generative AI
Full Conference Supporter
Full Conference