Requirements
Business requirements
The tool generates annotations, based on already marked elements, for the remaining elements of the set.
The tool has the ability to mark data and generate predictions if there are no base annotations, only a clean set.
The system should allow configuring model training parameters and selecting architectures to provide flexibility for different project needs.
Usage requirements
Possibility of continuing work with existing data sets.
Speeding up data annotation 10 times compared to humans.
System requirements
Functional:
The user can add projects by clicking a button, after clicking it, the project is entered into the database and the log is sent.
The user can add link to cloud storage where the data is stored.
The user can visually select data in the form of bounding boxes.
The user can assign a class to the selection.
The user can choose from pretrained/basic ML models.
The user can specify model hyperparameters by filling the form.
The user can see the results of the model in the form of numerical values and charts in the dashboard.
The user can download the generated annotations by pressing the appropriate button.
Non-functional:
It is possible to serve at least 2 clients.
It is possible to work with large datasets (more than 10000 images).
It is possible to store annotations in JSON format.