IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2022

Glass Segmentation using Intensity and Spectral Polarization Cues

Haiyang Mei¹ Bo Dong² Wen Dong¹ Jiaxi Yang¹ Seung-Hwan Baek^2,3 Felix Heide² Pieter Peers⁴ Xiaopeng Wei¹ Xin Yang^1,*

¹ Dalian University of Technology ² Princeton University ³ Pohang University of Science And Technology ⁴ College of William & Mary

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

RGB-P Glass Segmentation Dataset

1. Overview

We collected a large-scale polarization glass segmentation dataset, named RGBP-Glass using a LUCID PHX050S trichromatic polarizer-array camera that records four different linear-polarization directions (0, 45, 90, and 135 degree) for each color channel (i.e., R, G, and B). RGBP-Glass dataset contains 4511 RGB intensity and corresponding pixel-aligned trichromatic AoLP and DoLP images. All images come with manually annotated pixel-level accurate reference glass masks. Each image in RGBP-Glass contains at least one in-the-wild glass object. Our RGBP-Glass dataset offers rich diversities in scene and glass. Please refer to our paper for detailed statistics of the RGBP-Glass dataset.

2. File Structure and Dataset Distribution

As some glass segmentation methods require reflection and edge maps for training, the RGBP-Glass dataset also provides the ground truth associated to these additional information.

3. Downloads

Paper	: [ PGSNet.pdf ]
Experimental results	: [ Google Drive ] [ Baidu Disk, fetch code: rgbp ]
Pre-trained model	: [ Google Drive ] [ Baidu Disk, fetch code: rgbp ]
Backbone model	: [ Google Drive ] [ Baidu Disk, fetch code: rgbp ]
RGBP-Glass dataset	: [ Application Link ]
Code	: [ Github ]

4. BibTex

@InProceedings{Haiyang:PGSNet:2022,
    author = {Mei, Haiyang and Dong, Bo and Dong, Wen and Yang, Jiaxi and Baek, Seung-Hwan and Heide, Felix and Peers, Pieter and Wei, Xiaopeng and Yang, Xin},
    title = {Glass Segmentation using Intensity and Spectral Polarization Cues},
    booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month = {June},
    year = {2022}
}