autotrain-data-processor
Processed data from AutoTrain data processor ([2023-02-05 02:22 ]
913c95c
|
raw
history blame
954 Bytes
metadata
task_categories:
  - image-classification

AutoTrain Dataset for project: square-count-classifier

Dataset Description

This dataset has been automatically processed by AutoTrain for project square-count-classifier.

Languages

The BCP-47 code for the dataset's language is unk.

Dataset Structure

Data Instances

A sample from this dataset looks as follows:

[
  {
    "image": "<28x28 L PIL image>",
    "target": 0
  },
  {
    "image": "<28x28 L PIL image>",
    "target": 0
  }
]

Dataset Fields

The dataset has the following fields (also called "features"):

{
  "image": "Image(decode=True, id=None)",
  "target": "ClassLabel(names=['green', 'red'], id=None)"
}

Dataset Splits

This dataset is split into a train and validation split. The split sizes are as follow:

Split name Num samples
train 394
valid 40