The paper introduces VIP-SAM, a visual-instance-prompt segmentation method that identifies the specific garment from a flatlay image within a photograph of a person wearing it, rather than performing only category-level segmentation. It also presents CtrlVTON, which formulates virtual try-on as image editing and uses segmentation masks to control garment layout, including fit, styling, and spatial placement. According to the abstract, both components achieve state-of-the-art results on their respective tasks. CtrlVTON reportedly follows user-provided layouts more faithfully than leading proprietary editing systems while matching them in garment fidelity.
No heat snapshots are available in the last 24 hours.