datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
collected_pre_to_post_dataMET-Bench-Chess
MET-Bench: Multimodal Entity Tracking for Evaluating the Limitations of Vision-Language and Reasoning Models
Vanya Cohen and Raymond Mooney · ICML 2026
Paper · Publication page · Load the dataset · Citation
Domains: Chess · Shell Game · Minecraft
MET-Bench evaluates entity state tracking across text and image modalities. This repository contains the Chess domain.
Chess
Chess is an entity state tracking task in which a model follows the positions of pieces through… See the full description on the dataset page: https://huggingface.co/datasets/vanyacohen/MET-Bench-Chess.chess-openings
Dataset Card for Lichess Puzzles
Dataset Description
3704 chess openings with their Encyclopaedia of Chess Openings (ECO) classification.
Dataset Creation
Creating this dataset is detailed in its original GitHub repository.
Updates to the original repo will also be reflected in this version. Dataset last updated on May 20th, 2026.
Dataset Usage
Using the datasets library:
from datasets import load_dataset
dset =… See the full description on the dataset page: https://huggingface.co/datasets/Lichess/chess-openings.rs-all-objectsAssets from https://huggingface.co/datasets/nvidia/PhysicalAI-Robotics-Manipulation-Objects-Kitchen-MJCF
Please visit https://huggingface.co/datasets/nvidia/PhysicalAI-Robotics-Manipulation-Objects-Kitchen-MJCF for more information
chess-puzzles-images-mini
Dataset Card for Chess Puzzles Images (mini)
This dataset contains 124,999 chess board positions in JPG format, derived from Lichess puzzles. Each image is accompanied by a shortened FEN string, indication for the color to play as, castling and en passant availability, and best moves in standard algebraic notation.
The fields are as follows:
image: image, A visual representation of the chess board showing the current piece arrangement.
board_state: string, A shortened FEN… See the full description on the dataset page: https://huggingface.co/datasets/bingbangboom/chess-puzzles-images-mini.kaggle-chess-positions
Kaggle Chess Positions (mirror)
Public mirror of the Chess Positions dataset by koryakinp on Kaggle, hosted for easier access in ML pipelines (including 2d-chess-ocr).
Source & attribution
Field
Value
Original dataset
koryakinp/chess-positions
Original author
koryakinp
License
CC0-1.0
Generation tool
koryakinp/chess-generator
Board/piece assets
Chess.com
This repository is a mirror of the Kaggle release. Please cite the original Kaggle… See the full description on the dataset page: https://huggingface.co/datasets/smallchess/kaggle-chess-positions.chess-rfdetr-cocoChesschess-puzzles-images-large
Dataset Card for Chess Puzzles Images (large)
This dataset contains 1,249,999 chess board positions in JPG format, derived from Lichess puzzles. Each image is accompanied by a shortened FEN string, indication for the color to play as, castling and en passant availability, and best moves in standard algebraic notation.
The fields are as follows:
image: image, A visual representation of the chess board showing the current piece arrangement.
board_state: string, A shortened FEN… See the full description on the dataset page: https://huggingface.co/datasets/bingbangboom/chess-puzzles-images-large.chess-pieces-mjzgj
Dataset Card for chess-pieces-mjzgj
** The original COCO dataset is stored at dataset.tar.gz**
Dataset Summary
chess-pieces-mjzgj
Supported Tasks and Leaderboards
object-detection: The dataset can be used to train a model for Object Detection.
Languages
English
Dataset Structure
Data Instances
A data point comprises an image and its object annotations.
{
'image_id': 15,
'image': <PIL.JpegImagePlugin.JpegImageFile image… See the full description on the dataset page: https://huggingface.co/datasets/Francesco/chess-pieces-mjzgj.robot-chess-commentator-squares
Robot chess commentator: square dataset
Labelled square crops of a real chess board photographed by a Reachy Mini robot, labelled with the piece standing
on each square. 23,744 labelled squares from 371 board positions. Was used to train a square classification model that enables the robot to commentate chess games
Configs
squares -- one row per square crop
Split
Rows
train
13,184
val
4,928
test
5,632
Column
Notes… See the full description on the dataset page: https://huggingface.co/datasets/felixfabricius/robot-chess-commentator-squares.chessboard-dataset-yolo
Chessboard Corner Pose Dataset ♟️📐
1. Dataset Description
This dataset is designed for Keypoint Detection (Pose Estimation) tasks, specifically to detect the 4 semantic corners of a chessboard from various camera angles.
It is intended to be used as Phase 1 of a Chess Move Tracking pipeline. The detected keypoints are used to calculate the Homography matrix for Perspective Warping (straightening the board).
Task: Keypoint Detection (YOLO-Pose)
Format: YOLOv8 / YOLO11… See the full description on the dataset page: https://huggingface.co/datasets/surawut/chessboard-dataset-yolo.chess-board-to-fen
Chess Image To FEN Dataset
1M rows (80/20 train test split)
perfect split between opening, middle game and late game (33% of each)
Random variations applied to image (7 themes, board rotation and coordinates)
chess-pieces-dataset-yolo
Chess Pieces Detection Dataset (Warped View) ♟️🔍
1. Dataset Description
This dataset contains images of chess pieces taken from a Top-Down (Warped) perspective. It is designed for Object Detection models (YOLO) to recognize chess pieces and hands on a standardized 640x640 grid.
It is intended to be used as Phase 2 of a Chess Move Tracking pipeline, receiving input from the Phase 1 Perspective Warper.
Task: Object Detection
Format: YOLOv8 / YOLO11 Detection Format
Input:… See the full description on the dataset page: https://huggingface.co/datasets/surawut/chess-pieces-dataset-yolo.chess-gridfit-debugchess-pieces-dominique
Chess Piece Detection Dataset: chess_pieces_dominique
Dataset Description
This dataset contains chess piece detection annotations in YOLOv8 format.
Chess piece detection dataset from Dominique with 12 classes of chess pieces, optimized for YOLOv8 training.
Dataset Structure
The dataset follows the YOLOv8 format with the following structure:
train/: Training images and labels
valid/: Validation images and labels
test/: Test images and labels
Classes… See the full description on the dataset page: https://huggingface.co/datasets/dopaul/chess-pieces-dominique.chesssight-synthetic-40k
ChessSight synthetic boards (40k)
Photorealistic Blender renders of chess positions with exact labels: the position
as FEN and an 8×8 grid, the four board corners in image pixels, and an
occlusion-correct box per piece. Built to train board readers without hand
labelling, and validated by transfer to real photographs.
Licence and attribution — read this first
CC BY-NC 4.0. Commercial use is not permitted.
About 60% of these images render the Staunton Chess Set by… See the full description on the dataset page: https://huggingface.co/datasets/tchauffi/chesssight-synthetic-40k.old_chess_dataChessboard-digital-images_with_fen
Chessboard Detection Dataset
This dataset consists of a total of 64,386 chessboard images and corresponding YOLO-format label files.
Dataset Breakdown
Images: 64,386 total
train: 57,928
val: 6,458
Labels: 64,386 total (one .txt per image)
train: 57,928
val: 6,458
Each label file contains bounding boxes for the pieces on the board using YOLO format. The dataset includes 12 classes:
6 white pieces
6 black pieces
Data Collection & Annotation… See the full description on the dataset page: https://huggingface.co/datasets/MohammedHemed/Chessboard-digital-images_with_fen.chess-openings
Dataset Card for Lichess Puzzles
Dataset Description
3704 chess openings with their Encyclopaedia of Chess Openings (ECO) classification.
Dataset Creation
Creating this dataset is detailed in its original GitHub repository.
Updates to the original repo will also be reflected in this version. Dataset last updated on May 20th, 2026.
Dataset Usage
Using the datasets library:
from datasets import load_dataset
dset =… See the full description on the dataset page: https://huggingface.co/datasets/xxx3467/chess-openings.chess-pieces-roboflow
Chess Piece Detection Dataset: chess_pieces_roboflow
Dataset Description
This dataset contains chess piece detection annotations in YOLOv8 format.
Chess piece detection dataset from Roboflow with processed labels, cleaned and standardized for YOLOv8 format.
Dataset Structure
The dataset follows the YOLOv8 format with the following structure:
train/: Training images and labels
valid/: Validation images and labels
test/: Test images and labels
Classes… See the full description on the dataset page: https://huggingface.co/datasets/dopaul/chess-pieces-roboflow.chess-pieces-merged
Chess Piece Detection Datasets: merged-chess_pieces_dominique-chess_pieces_roboflow
Dataset Description
This is a merged dataset combining multiple chess piece detection datasets.
Comprehensive chess piece detection dataset combining multiple high-quality sources. This merged dataset provides more training data and better generalization for chess piece detection models.
Dataset Structure
The dataset follows the YOLOv8 format with the following structure:… See the full description on the dataset page: https://huggingface.co/datasets/dopaul/chess-pieces-merged.chess-puzzle-tasks-visionchess_piece_and_empty_square_trainingInformation about the dataset:
Pieces: 11 pieces(white & black) -> 32 variations -> 1000 photos for each variation 352,000 photos
Board: 18 boards -> 5000 photos of empty squares including all variations
Information about accessing the data set
python
from datasets import load_dataset
ds = load_dataset("HardlySalty/chess_piece_and_empty_square_training")
print(ds["train"].features["label"].names)
chess_pieceschess_checkmate_images_big_bench
Dataset Card for BIG-Bench Checkmate In One Move (Images)
Description
This is an adapted version of the BIG-Bench Checkmate in One Move task
as originally made by Nitish Keskar (nkeskar@salesforce.com).
Copying the original task description:
The goal of this task is to probe the ability of language models to play chess in standard algebraic notation (SAN). The input to the model is a sequence of moves such that a next possible move is a checkmate. We curate 3,500 games… See the full description on the dataset page: https://huggingface.co/datasets/DylanASHillier/chess_checkmate_images_big_bench.chessai-data
ChessAI Data — Chinese Chess Piece Detection
A bounding-box detection dataset for Chinese Chess (Cờ tướng / 象棋), labeled with AnyLabeling. Built to train piece-recognition models for chessboard-state extraction from photos.
Dataset summary
Total annotated images: 1,747 (per-image AnyLabeling JSON in data/combined_data/)
COCO-format split: 872 images, 18,790 bounding boxes (data/annotations.json)
Classes: 7 (the standard Xiangqi piece set)
Total size: ~280 MB (images +… See the full description on the dataset page: https://huggingface.co/datasets/vietanhdev/chessai-data.chess-cv-openboard
OpenBoard Dataset (Piece Classification)
This dataset is a modified and simplified version of the OpenBoard-Dataset by Szustarol.
Dataset Description
This dataset contains images of chess pieces for classification. It is derived from the original OpenBoard-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… See the full description on the dataset page: https://huggingface.co/datasets/S1M0N38/chess-cv-openboard.HCS_Dataset-csvchess-positions-cv
Chess Positions — CV-derived annotations (cells + YOLO boxes)
Annotations and a runnable sample derived from the synthetic, CC0
koryakinp/chess-positions
dataset (400x400 rendered boards, FEN in filename, clean 8x8 grid → 50x50 cells).
This repo adds, on top of that source:
train_manifest.json / test_manifest.json — per-board records:
{path, fen, cell_labels[64], boxes, n_pieces}. cell_labels are 13-class CNN
indices (0=empty, 1-6 white PNBRQK, 7-12 black pnbrqk); boxes are… See the full description on the dataset page: https://huggingface.co/datasets/honi05/chess-positions-cv.