Materialistic: Selecting Similar Materials in Images
نویسندگان
چکیده
Separating an image into meaningful underlying components is a crucial first step for both editing and understanding images. We present method capable of selecting the regions photograph exhibiting same material as artist-chosen area. Our proposed approach robust to shading, specular highlights, cast shadows, enabling selection in real As we do not rely on semantic segmentation (different woods or metal should be selected together), formulate problem similarity-based grouping based user-provided location. In particular, propose leverage unsupervised DINO [Caron et al. 2021] features coupled with Cross-Similarity Feature Weighting module MLP head extract similarities image. train our model new synthetic dataset, that release. show generalizes well real-world carefully analyze model's behavior varying properties lighting. Additionally, evaluate it against hand-annotated benchmark 50 photographs. further demonstrate set applications, including editing, in-video selection, retrieval object photographs similar materials.
منابع مشابه
Spontaneous tribocharging of similar materials
We investigate the spontaneous triboelectrification of similar materials. This effect, first reported in 1927, has been little studied but is easily reproduced. We find in two separate experimental systems, where materials are prepared in the same way and rubbed symmetrically, that symmetry breaking occurs so that one material becomes positive and the other negative. Curiously, the distribution...
متن کاملSelecting Materials for Radiation Balanced Lasers
Well-characterized solid-state laser materials are evaluated for performance in radiation balanced laser systems. New figures-of-merit are developed and applied to ytterbium doped materials. Superior performance is predicted for high cross section tungstate materials. Photothermal deflection experiments on samples of Yb doped KGd(WO4)2 confirm anti-Stokes fluorescence cooling. Observations of o...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: ACM Transactions on Graphics
سال: 2023
ISSN: ['0730-0301', '1557-7368']
DOI: https://doi.org/10.1145/3592390