COOL: Contour and Object-Oriented Learning for Indoor Cleanliness Classification

Abstract

The evaluation of indoor cleanliness is a meaningful task for vision-based household service systems. However, the perception of cleanliness is determined by diverse visual features and multiple criteria, which is subjective to the observer. We find the existing dataset and method fail to truthfully capture the concept of cleanliness because the feature used is not representative to human subjective judgement. Therefore, we create a dataset for indoor cleanliness classification from a group of annotators based on SUN-RGBD, a richly annotated scene understanding benchmark. Based on such analysis, we propose Contour and Object-oriented Learning (COOL) model that integrates pretrained convolutional feature, low-level contour feature, and object arrangement in order to truthfully model the notion of cleanliness. Our design choices are justified in ablation studies, and our model outperforms the previous method in our dataset for cleanliness classification.

Authors

 
Sucheng Qian Zhaoyu Li Weibang Jiang

Demo

An introduction video to this project can be downloaded here link.

Paper

Code

Prerequisites

Getting Started

Citation

@article{Qian2019COOL
    author = {Sucheng, Qian and Zhaoyu, Li and Weibang, Jiang",
    title = {Contour and Object-Oriented Learning for Indoor Cleanliness Classification},
    year = {2019},
    howpublished={\url{https://github.com/OolongQian/Cleanliess-Classification}}
}

Acknowledgements

references

[1]  @inproceedings{Xie_ICCV_2015,
         author = {Saining Xie and Zhuowen Tu},
         title = {Holistically-Nested Edge Detection},
         booktitle = {IEEE International Conference on Computer Vision},
         year = {2015}
     }
[2]  @misc{pytorch-hed,
         author = {Simon Niklaus},
         title = {A Reimplementation of {HED} Using {PyTorch}},
         year = {2018},
         howpublished = {\url{https://github.com/sniklaus/pytorch-hed}}
    }