supervisely_lib

Supervisely Python SDK

This SDK aims to make it as easy as possible to develop new plugins for the Supervisely platform. Check out also our guide on how to package your plugin as a Docker image and make it available in the platform.

The SDK is a product of our experience developind Supervisely plugins and contains functionality that we have found helpful and frequently needed for pluginc development. The key areas covered by the SDK are

• Reading, modifying and writing Supervisely projects on disk.
• Working with labeling data: geometric objects and tags.
• Common functionality for developing Supervisely plugins, such as neural network models.

The detailed documentation is still under active development. Here is a broad outline of the available modules and areas they cover:

• annnotation - Working with labeling data of individual images. Annotation is the class that wraps all the labeling data for a given image: its Labels (geometrical objects) and Tags. See this tutorial on how to work with annotations in detail.
• api - Python wrappers to script your interactions with the Supervisely web instance. Instead of clicking around, you can write a script to request, via the API, a sequence of tasks, like training up a neural network and then running inference on a validation dataset.
• aug - Data augmentations to create more data variety for neural networks training.
• geometry - All the logic concerned with working with geometric objects - compute statistics like object area, transform (rotate, scale, shift), extract bounding boxes, compute intersections and more.
• imaging - Our wrappers for working with images. IO, transformations, text rendering, color conversions.
• io - Low-level convenience IO wrappers that we found useful internally.
• metric - Computing similarity metrics between annotations. Precision, recall, intersection over union, mean average precision etc. Also includes wrappers to easily package metric computations as Supervisely plugins.
• nn - All the common code for working with neural networks. Includes a class to easily iterate over a Supervisely dataset during neural network training. Also conversions between raw tensors and Supervisely labeling data.
  • hosted - All the neural network logic that is specific to interacting with Supervisely platform. Base classes for neural network training and inference with common functionality already implemented (see our guide on adding your own neural network plugin), parsing, interpreting and consistency checking of configs, class indexing (between integer IDs of the tensors and Supervisely classes) helpers.
  • pytorch - Convenience wrappers for PyTorch - model weights IO, metrics etc.
  • training - A helper to flexibly schedule when to run validation during the model training.
• project - Working with Superrvisely projects on disk.
• task - Constants defining the directories where a plugin should expect the input and output data to be. Also helpers to stream progress data from a running plugin back to the web instance.