Where Is My Mirror?
Xin Yang1,* Haiyang Mei1,* Ke Xu1,3 Xiaopeng Wei1 Baocai Yin1,2 Rynson W.H. Lau3,†1 Dalian University of Technology 2 Peng Cheng Laboratory 3 City University of Hong Kong
* Joint first authors † Rynson Lau is the corresponding author and he led this project
Mirrors are everywhere in our daily lives. Existing computer vision systems do not consider mirrors, and hence may get confused by the reflected content inside a mirror, resulting in a severe performance degradation. However, separating the real content outside a mirror from the reflected content inside it is non-trivial. The key challenge lies in that mirrors typically reflect contents similar to their surroundings, making it very difficult to differentiate the two. In this paper, we present a novel method to accurately segment mirrors from an input image. To the best of our knowledge, this is the first work to address the mirror segmentation problem with a computational approach. We make the following contributions. First, we construct a large-scale mirror dataset that contains mirror images with the corresponding manually annotated masks. This dataset covers a variety of daily life scenes, and will be made publicly available for future research. Second, we propose a novel network, called MirrorNet, for mirror segmentation, by modeling both semantical and low-level color/texture discontinuities between the contents inside and outside of the mirrors. Third, we conduct extensive experiments to evaluate the proposed method, and show that it outperforms the carefully chosen baselines from the state-of-the-art detection and segmentation methods.
Mirror Segmentation Dataset (MSD)
Updated Dataset Statistics
|(a) mirror area distribution||(b) mirror shape distribution||(c) mirror location distribution||(d) color contrast distribution|
|Paper||: [ MirrorNet.pdf ]|
|Mirror Segmenation Dataset||: [ Google Drive ]|
|Experimental results||: [ Results.zip ]|
|Pre-trained model.||: [ MirrorNet.pth ]|
|Source Code.||: [ Code ]|