---
license: cc-by-nc-3.0
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
dataset_info:
  features:
  - name: image
    dtype: image
  - name: label
    dtype:
      class_label:
        names:
          '0': bB
          '1': bK
          '2': bN
          '3': bP
          '4': bQ
          '5': bR
          '6': wB
          '7': wK
          '8': wN
          '9': wP
          '10': wQ
          '11': wR
          '12': xx
  splits:
  - name: train
    num_bytes: 5663395
    num_examples: 2591
  - name: validation
    num_bytes: 1249307
    num_examples: 595
  download_size: 6939131
  dataset_size: 6912702
task_categories:
- image-classification
tags:
- chess
- cnn
- chess-cv
size_categories:
- 1K<n<10K
---

# ChessVision Dataset (Piece Classification)

This dataset is derived from the [ChessVision-3LC](https://github.com/gudbrandtandberg/ChessVision-3LC) dataset by Gudbrand Tandberg, adapted for chess piece classification.

## Dataset Description

This dataset contains images of chess pieces for classification. It is derived from the original ChessVision-3LC dataset and has the following modifications:

- **Purpose:** The dataset is intended for training and testing image classification models to recognize chess pieces.
- **Structure:** The images are organized in an ImageFolder structure, ready to be used with popular deep learning frameworks. The directories are named after the piece they contain (e.g., `wP` for white pawn, `bK` for black king, `xx` for empty squares).
- **Image Format:** All images are 32x32 pixel PNG files (RGB format), optimized for storage and fast loading.

For more information about the original data collection process, please refer to the [original repository](https://github.com/gudbrandtandberg/ChessVision-3LC).

## Categories

The dataset contains 13 categories:

- **White pieces:** `wR` (Rook), `wN` (Knight), `wB` (Bishop), `wQ` (Queen), `wK` (King), `wP` (Pawn)
- **Black pieces:** `bR` (Rook), `bN` (Knight), `bB` (Bishop), `bQ` (Queen), `bK` (King), `bP` (Pawn)
- **Empty square:** `xx` (no piece on the square)

## Dataset Structure

The dataset is split into two subsets:

- **Train:** Training images
- **Validation:** Validation images

Each subset contains images organized by category in subdirectories.