Dice Recognition Datasets for Machine Learning

サイコロの目認識のための教師用画像データセット

Introduction

MNIST(LeCun 1998) is a digital handwritten image dataset. This is a very famous image dataset used for benchmarking machine learning. We have created a new image dataset that can be used to practice machine learning for classification such as MNIST. It is an image data set of dice that everyone knows. Image datasets are created and annotated using Dice Recognition System.

画像認識（分類）のための教師用画像データセットとして手書き数字のMNISTが有名です。ちょっと変わったサイコロの目の画像データセットを準備しました。サイコロの目の認識などにいかがでしょうか？ご自由にお使いください。

グレースケールサイコロ画像データセットはこのシステムでサイコロの目を認識し、画像データセットを作成しました。 色付きサイコロ画像データセットは教師画像データ作成システムで、画像データセットを作成しました。

Grayscale dice image dataset

• Dataset Download

Download

Detail of Grayscale dice image dataset

We used two different size and color of dice. / サイズと色が異なる2種類のサイコロを使用しています。

• Image

category	detail
Image Format	BMP
size	128 x 128
channel	1 (Gray scale)
depth	8

• labeld dataset

Labeld Dice	count
1	400
2	400
3	400
4	400
5	400
6	400

Colored dice image dataset

• Dataset Download

Download

Detail of Colored dice image dataset

We used different size and color of dice. / サイズと色が異なるサイコロを使用しています。

• Image

category	detail
Image Format	JPG
size	256 x 256
channel	3 (Color)
depth	8

• labeld dataset

Labeld Dice	count
1	324
2	324
3	324
4	324
5	324
6	324

Other

Exhibition Information

We exhibited this project at NT加賀2018 for the first time.

A sequel to the NT加賀2018

@うえぽん created a dice classification model using deep learning. However, the image data captured from the real world had a low recognition rate due to the effects of fluorescent lights flicker and shadows. We thought that augmentation for deep learning and in addition learning according to the real world was necessary. I think that it is domain knowledge or learning in HW on the real world side.

Source code for training using DNN

The source code for training this dataset is uploaded on this site.(Author llDataSciencell)

• https://github.com/llDataSciencell/DiceRecognitionKeras/

Dice Recognition Challenge

Name	Remark
@tomit3	Recognize the face of dice by counting circles using image processing and Hough transform. Co-author of this project.
@うえぽん/llDataSciencell	He is the first person in the world who made the model of dice recognition using Deep Learning. The training code using DNN is written below. Co-author of this project.
@ミクミンP/Kazuhiro Sasao	Recognized the face of dice by transfer learning. He has written an article on transfer learning techniques in an offline using Windows. Articale is "Windows PCかつオフライン環境でお気軽に画像認識する"