Creating a new split
Layout for new split creation

Parameters used for creating a new split

  • Split name
  • Type of model would you like to experiment with - It includes Attribute prediction, Label class prediction, Object detection, Image tag prediction, Instance segmentation, Semantic segmentation.
  • Split strategy - The user has the option to choose a data split strategy. It includes Random, Stratification, Dataset partioning.
Random sampling is the simplest split strategy and is appropriate in most cases.
Stratification may help if the dataset has just a few samples of particular label class sets.
Dataset partitioning is appropriate when the data split has already been performed and the partitions have been uploaded in separate datasets.
  • Part of a project to be used for split creation
Available image status types which can be used
  • Datasets - This gives the user the flexibility to select the datasets which are required for a particular model building experiment.
  • Split percentage ratio - The user can tweak the percentage of images to be divided into the Train, Test and Validation datasets
The default value of percentage split for the Train, Test and Validation datasets are 70%, 20% and 10% respectively.