IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2020

Don't Hit Me! Glass Detection in Real-world Scenes

Haiyang Mei¹ Xin Yang^1,4,* Yang Wang¹ Yuanyuan Liu¹ Shengfeng He²
Qiang Zhang¹ Xiaopeng Wei^1,* Rynson W.H. Lau³

¹ Dalian University of Technology     ² South China University of Technology
³ City University of Hong Kong     ⁴ Advanced Institute of Information Technology Peking University

Contact us:    xinyang@dlut.edu.cn    mhy666@mail.dlut.edu.cn

Abstract

Glass is very common in our daily life. Existing computer vision systems neglect the glass and thus might lead to severe consequence, e.g., the robot might crash into the glass wall. However, sensing the presence of the glass is not straightforward. The key challenge is that arbitrary objects/scenes can appear behind the glass and the content presented in the glass region typically similar to those outside of it. In this paper, we raise an interesting but important problem of detecting glass from a single RGB image. To address this problem, we construct a large-scale glass detection dataset (GDD) and design a glass detection network, called GDNet, by learning abundant contextual features from a global perspective with a novel large-field contextual feature integration module. Extensive experiments demonstrate the proposed method achieves superior glass detection results on our GDD test set. Particularly, we outperform state-of-the-art methods that fine-tuned for glass detection.

GDNet

Visual Results

Downloads

Paper	: [ GDNet.pdf ]
Experimental results	: [ Results.zip ]
Pre-trained model	: [ GDNet.pth ]
Source Code	: [ Code ]
Dataset	: [ Application Link ]

BibTex

@InProceedings{Mei_2020_CVPR,
    author = {Mei, Haiyang and Yang, Xin and Wang, Yang and Liu, Yuanyuan and He, Shengfeng and Zhang, Qiang and Wei, Xiaopeng and Lau, Rynson W.H.},
    title = {Don't Hit Me! Glass Detection in Real-World Scenes},
    booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month = {June},
    year = {2020}
}