This repository makes available the public dataset for the hand, which has 6720 image samples. These images were captured by 15 individuals, 9 women and 6 men, in different simple backgrounds. Other variations of data such as light condition, different capturing devise (mobile and laptop), and different resolutions are considered. The presented dataset consists of 4 classes which are:

• Fist
• Palm
• Left
• Right

Some of the training samples for each class:

Following links let you download training and validation set.

Training Set (79 MB)

Validation Set (9 MB)

Please notice that training set and validation set have different distribution.

Each sample in training set is a 100 × 100 x 3 image associated with a label from 4 classes. Validation data were captured by SSD Hand Detection; Therefore, sample in validation set have different size but 3 dimensions.

In my case, the dataset was classified with EfficientNet-B0 and some fully-connected layers in order to control pointer of computer with hand detection and classification. After downloading dataset, you can use load dataset.py to unzip 6720 hand images and make your training data and its labels.