metadata
license: mit
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
dataset_info:
features:
- name: image
dtype: image
- name: label
dtype:
class_label:
names:
'0': apples
'1': bananas
'2': bottles
'3': cans
'4': cardboard
'5': cups
'6': eggshells
'7': generalcompost
'8': mixers
'9': peels
'10': plasticbags
'11': plastics
'12': tissues
splits:
- name: train
num_bytes: 122444841
num_examples: 14651
download_size: 2050293304
dataset_size: 122444841
The dataset has images collected from publicly available resources like Kaggle and Roboflow, and some photos that I clicked.
Feel free to expand on the ones available and add more directories.
To get an idea of which additional directories could be useful refer recycle.jpeg and compost.jpeg.
The notebook used to train the dataset and the best performing model with 98.2947% accuracy is saved at https://huggingface.co/dvk65/trash-classifier-resnet50.
To use this dataset in your python project use:
from datasets import load_dataset
dataset = load_dataset("dvk65/TrashTypes", split="train")
label_names = dataset.features["label"].names
Currently, it is in a single train split.