Published on:
5 December 2023
Primary Category:
Computer Vision and Pattern Recognition
Paper Authors:
Prafull Sharma,
Varun Jampani,
Yuanzhen Li,
Xuhui Jia,
Dmitry Lagun,
Fredo Durand,
William T. Freeman,
Mark Matthews
Proposes method to control material properties in images
Renders synthetic dataset with explicit material annotations
Fine-tunes text-to-image model on this dataset
Edits roughness, metallic, albedo and transparency
Works directly in pixel space without graphics pipelines
Can also edit materials in neural radiance fields
Controlling Material Properties in Images
This paper proposes a method to precisely control key material properties like roughness, metallic, albedo and transparency in real images. It uses a text-to-image model fine-tuned on a synthetic dataset with explicit material annotations. This allows editing materials in images while retaining all other attributes, without needing to estimate geometry, lighting etc. They show it can also edit neural radiance fields.
No comments yet, be the first to start the conversation...
Sign up to comment on this paper