metadata

task_categories:
  - image-classification
language:
  - en
tags:
  - Images
pretty_name: 'Material Classification Hands On '
size_categories:
  - n<1K
dataset_info:
  config_name: plain_text
  features:
    - name: image
      dtype: image
    - name: label
      dtype:
        class_label:
          names:
            '0': Brick
            '1': Metal
            '2': Paper
            '3': Plastic
            '4': Wood
  splits:
    - name: train
      num_examples: 120
    - name: test
      num_examples: 30
license: mit

Dataset Card for Material Classification

Dataset Description

Homepage: https://semillerocv.github.io/proyectos.html
Repository: https://github.com/Sneider-exe/Clasificacion_Materiales

Dataset Summary

The Material_classification_2U dataset consists of 150 256x256 color images, categorized into 5 classes with 30 images per class. The dataset is divided into two main subsets: 120 images for training and 30 images for testing. Each image is labeled into one of the following five categories: Brick, Metal, Paper, Plastic, and Wood.

Supported Tasks and Leaderboards

image-classification: The goal of this task is to classify a given image into one of 5 classes.

Languages

English

Dataset Structure

Data Instances

A sample from the training set is provided below:

{
  'image': <PIL.PngImagePlugin.PngImageFile image mode=RGB size=256x256>,
  'label': 1
}

Data Fields

image: A PIL.Image.Image object containing the 256x256 image. Note that when accessing the image column: dataset['train']["image"] the image file is automatically decoded. Decoding of a large number of image files might take a significant amount of time.
label: 0-4 with the following correspondence '0': Brick '1': Metal '2': Paper '3': Plastic '4': Wood

Data Splits

The dataset is divided into two main subsets: Train and Test.

Train Split:
- Number of Images: 120
- Distribution: 24 images per class
Test Split:
- Number of Images: 30
- Distribution: 6 images per class

Both splits are stratified, ensuring that each class is proportionally represented in both the Train and Test subsets. This means that the percentage of images for each class remains consistent across both splits, providing a balanced and representative distribution for model training and evaluation.

Citation Information

@TECHREPORT{
    author = {Brayan Sneider Sánchez, Dana Meliza Villamizar, Cesar Vanegas, Juan Jose Calderón},
    title = {Material Classification},
    institution = {Universidad Industrial de Santander},
    year = {2024}
}