Chess Checkmate Images
This dataset contains 3,500 chess board images taken from real games, each representing the state of the board just before a checkmate hit. It is a visual version of the BIG-Bench benchmark, useful for training or testing AI models in recognizing visual patterns in the game of chess.
Description
The dataset Chess Checkmate Images is a visual adaptation of a classic chess reasoning benchmark. It contains 3,500 PNG images representing chess positions just before a checkmate. Each image is associated with the player's turn (white or black) and the expected solution (checkmate).
What is this dataset for?
- Testing the ability of computer vision models to understand the dynamics of the game of chess
- Train multimodal models combining vision and logical reasoning
- Create intelligent agents that can play or analyze chess games from images
Can it be enriched or improved?
Yes, you can enrich the dataset with metadata such as the complete game in PGN, the classification of the types of checkmate (double check, one-hit, etc.), or annotations by expert player. It is also possible to adapt the visuals to vary the styles of pieces and chessboards.
🔎 In summary
🧠 Recommended for
- Computer vision developers
- Visual reasoning researchers
- Smart game creators
🔧 Compatible tools
- OpenCV
- PyTorch
- TensorFlow
- Keras
💡 Tip
Use binary classification (mat found/not found) to train lightweight models effectively.
Frequently Asked Questions
Is this dataset adapted to classical vision models?
Yes, it can be used with any image classification model like ResNet, VGG, or ConvNext.
Is there an annotation to indicate the correct curse?
Yes, each frame is associated with an expected hit of the mat, according to standard chess notation.
Can it be used to train a visual chess engine?
Yes, it's even a great starting point for training or testing a chess engine based on visual recognition.
