Neural Scene Decoration from a Single Photograph

Furnishing and rendering indoor scenes has been a long-standing task for interior design, where artists create a conceptual design for a space, build a 3D model, decorate it, and then render the result. Although important, the pipeline is tedious and requires tremendous effort.

This work introduces neural scene decoration: given a photograph of an empty indoor space and a list of decorations with a user-determined layout, synthesize a new image of the same space with the desired furnishing. The proposed scene-generation architecture transforms an empty scene and an object layout into a realistic furnished scene photograph, and is validated against conditional image-synthesis baselines both qualitatively and quantitatively.

← All projects