Dataset Viewer
Auto-converted to Parquet
Search is not available for this dataset
image
imagewidth (px)
27
170
label
class label
34 classes
01b
01b
01b
01b
01b
01b
01b
01b
01b
01b
01b
01b
01b
01b
01b
01b
01b
01b
01b
01b
01b
01b
01b
01b
01b
11n
11n
11n
11n
11n
11n
11n
11n
11n
11n
11n
11n
11n
11n
11n
11n
11n
11n
11n
11n
11n
11n
11n
11n
11n
11n
11n
21p
21p
21p
21p
21p
21p
21p
21p
21p
21p
21p
21p
21p
21p
21p
21p
21p
21p
21p
21p
32b
32b
32b
32b
32b
32b
32b
32b
32b
32b
32b
32b
32b
32b
32b
32b
32b
32b
32b
32b
32b
32b
32b
32b
32b
32b
32b
32b
End of preview. Expand in Data Studio

Dataset Card for Mahjong Souls Tiles

Dataset Details

Dataset Description

This dataset, pjura/mahjong_souls_tiles, contains images of Mahjong tiles extracted from the game Mahjong Soul. It was created specifically for the purpose of training computer vision models to recognize Mahjong tiles within the game interface.

The dataset was used to fine-tune the google/vit-base-patch16-224-in21k model, resulting in the pjura/mahjong_vision model available on Hugging Face.

  • Curated by: pjura
  • License: apache-2.0

Dataset Sources

Uses

Direct Use

This dataset is primarily intended for training or fine-tuning image classification models, specifically Vision Transformers (ViT) or similar architectures, for the task of recognizing Mahjong tiles from game screenshots or live feeds.

Out-of-Scope Use

Using this dataset for tasks unrelated to Mahjong tile recognition might not yield meaningful results. The images are specific to the art style and presentation of the Mahjong Soul game.

Dataset Structure

The dataset consists of labeled images of individual Mahjong tiles. It is structured as an image folder dataset suitable for use with libraries like datasets and torchvision.

Associated Model Performance

The pjura/mahjong_vision model, fine-tuned on this dataset, achieved the following results on the evaluation set:

  • Loss: 0.0466
  • Accuracy: 0.9967
  • F1: 0.9966
  • Recall: 0.9967

Associated Model Training Details

Training Hyperparameters

The pjura/mahjong_vision model was trained with the following hyperparameters:

  • learning_rate: 5e-05
  • train_batch_size: 16
  • eval_batch_size: 16
  • seed: 42
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 64
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_ratio: 0.1
  • num_epochs: 250

Framework Versions

  • Transformers 4.34.1
  • Pytorch 2.1.0+cu121
  • Datasets 2.14.6
  • Tokenizers 0.14.1

Citation

@misc{mahjong_souls_tiles_dataset,
  author = {pjura},
  title = {Mahjong Souls Tiles Dataset},
  year = {2023},
  publisher = {Hugging Face},
  journal = {Hugging Face repository},
  howpublished = {\url{https://huggingface.co/datasets/pjura/mahjong_souls_tiles}}
}

Dataset Card Contact

pjura on Hugging Face (https://huggingface.co/pjura)

Downloads last month
95

Models trained or fine-tuned on pjura/mahjong_souls_tiles