of 1380 masks + originals. In each image, the hands and each finger of the hand are segmented separately. This dataset will help train your model to understand the position of the hands and fingers in space.
The use of the dataset for commercial purposes is allowed only with the consent of the copyright holder (manager@labelme.ru )

The task of semantic segmentation is more complex than the task of image classification and object search, which is due not only to the need to determine the classes of objects, but also to identify their structure, the correct allocation of parts of objects in the image.

Color map: Background - black (0, 0, 0); Palm - (127, 127, 255); Thumb - (206, 36, 68); Index finger - (127, 255, 255); Middle finge - pink (255, 127, 255); Ring finger - (255, 255, 127); Little finger - white (255, 255, 255). The dataset contains 1380 fully segmented masks and exactly the same number of originals.

You can read more about this dataset in our first article: Segmentation of the hand from scratch: tells the manager for working with markup.