Waste Classification Dataset
Description
The dataset provides a collection of 24,705 images of solid household waste, categorised into two classes: organic (13,880) and recyclable (10,825) . The data is a restructured and represented version of the original dataset by Sashaank Sekar available at https://www.kaggle.com/techsash/waste-classification-data. The original dataset from Kaggle consists of 25,077 images of organic (13,966) and recyclable (11,111) images. However, performed som clean-up operations explained in the "further note" to reduce the data to 24,705 with (13,880 organic) and (10,825 recyclable) The restructured data has been used in a research study undertaken by academics and researchers at Computer Science Department, Edge Hill University, United Kingdom. To encourage reproducibility of the experiments and results reported, the modified data; a Jupyter notebook (.ipynb) file useful to apply data augmentation on the dataset; and a Jupyter notebook (.ipynb) file useful to replicate the experiments has been provided.
Files
Steps to reproduce
The original dataset from Kaggle consists of 25,077 images of organic (13,966) and recyclable (11,111) images. The acquired images are coloured .jpg files of randomly portrait and landscape orientation with resolution ranging from 191 pixels (minimum) x 264 pixels (maximum). A total of 24,705 images have RGB colour mode while 372 images have P mode. The latter was removed from the dataset to avoid colour-banding issues. Thus, only 24,705 images was retained for experiments. To enhance the size and quality of the dataset, we applied data augmentation which includes a suite of techniques for increasing the amount of data by adding slightly modified copies of the original data. The experimental data is included here (without augmentation) a Jupyter notebook file (.ipynb) file is provided which contains code used to perform augmentation.