Image Dataset of Domestic Organic Waste and Non-Organic Contaminants for Classification and Segmentation

Image Dataset of Domestic Organic Waste and Non-Organic Contaminants for Classification and Segmentation

We developed a Smart Trash Can that allows to unobtrusively take photos of real waste in the producer's home. Over six months, a volunteering family collected a total of 450 raw photos of domestic organic waste and non-organic contaminants, which were then manually labeled and segmented according to the captured waste types. Of the total of 450 images, 119 show pure organic waste while 324 also captured intentionally added contaminants. Another 7 photos show only the side walls or the plastic bag without any added waste, and are hence considered as background.

Context and Methodology

• Image dataset for the training of machine learning-models for computer vision aiming at waste quality management in the producer's home
• Collection of domestic organic waste and non-organic contaminants with a Smart Trash Can
• Images manually labeled and segmented according to the waste types: organic, non-organic, and background
• The dataset was used in a publication demonstrating two machine-learning approaches for computer vision, a binary classification (mean accuracy of 90.35 %) and a impurity segmentation (mean accuracy of 98.24 %, mean intersection over union value of 96.43 %)

Technical Details

• A total of 450 photos: 119 pure organic, 324 with intentionally added contaminants, 7 background
• Domestic waste of a volunteering family (3 male, 2 female; 15-55 yrs; 33.2 ±16.2 yrs)
• Subjects gave written consent to provide the image data for research purposes and publication
• Raw photos provided in a lossless *.png format with compression level 0
• Two folders 'images' and 'masks' containing the *.png images and the *.png masks for semantic segmentation, respectively
• The images are classified and grouped in the three sub-folders 'background', 'organic', and 'non-organic'
• The semantic segmentation labels associated with the colors in the masks are provided in the 'labelmap.csv' file (';' as separator)