Skip to content
master
Switch branches/tags
Code

Latest commit

[agent] disable cache app sources for custom branch
[agent] increase shm_size for docker containers

[SDK] sly project to detection / segementation / instance segmentation tasks
[SDK] add optional imageId to annotation
[SDK] tags in ImageInfo (downalod images tags without downloading entire annotation)
[SDK] fix sly.imguag_utils.aug_to_python
[SDK] sly.imguag_utils.apply_to_image
[SDK] add progress to sly.fs.download by url
[SDK] sly.project.download with image info (disabled by default)
[SDK] add min_report_percent to progress

[API] api.file.upload_directory with progress bar
[API] api.project.download_images_tags (tags to iamges dict)

[App Engine] compile template from multiple files
[App Engine] scrollIntoView (automatic initialization)
[App Engine] ignore_errors_and_show_dialog_window decorator
[Widget SDK] CompareGallery
047e52e

Git stats

Files

Permalink
Failed to load latest commit information.


Supervisely
Supervisely

AI for everyone! Neural networks, tools and Python SDK we use with Supervisely.

IntroductionAgentNeural NetworksImport PluginsPython SDKData TransformationsResources

screenshot

Introduction

Supervisely is a web platform where you can find everything you need to build Deep Learning solutions within a single environment.

We learn a lot from our awesome community and want to give something back. Here you can find the Python SDK we use to implement neural network models, import plugins for custom data formats, and tools like the Data Transformation Language. You can also find the source code for the agent to turn your PC into a worker node to deploy your neural nets.

Agent

Supervisely Agent is a simple open-source task manager available as a Docker image.

The Agent runs on a worker node. It connects to the Supervisely WEB instance and listens for the tasks (like neural net training request) to run. It handles downloading and uploading data to the web instance, sets up proper environments for the specific tasks to run, and keeps track of progress, successes and failures of individual tasks. By deploying the agent on worker machine you bring up a virtual computing claster that your team can run their tasks on from the Supervisely web instance.

Deploying agent to Supervisely

Check out explanation on how agent works and documentation on how to deploy a new agent.

Neural Networks

We have ported and implemented a number of popular neural network (NN) architectures to be available as Supervisely plugins. Each plugin is a separate Docker image. We continue to work on porting more NN architectures to Supervisely.

We also have a detailed guide on how to make your own neural net plugin, so you do not have to depend on anyone else to be able to use your favorite architecture.

Here are the neural network architectures available out of the box today:

Read here how to run training or inference with this models.

For all source implementations of NNs the original authors are retaining all their original rights.

Import plugins

We provide import plugins for over 30 popular dataset formats out of the box. You can also leverage our Python SDK to implement a new plugin for your custom format.

Python SDK

We have organized most of the common functionality for processing data in Supervisely format and for training and inference with neural networks into the Python SDK. Our stock plugins rely on the SDK extensively, and we believe the SDK will be also valuable to the community.

The SDK not only wraps all the low-level details of handling the data and communicating with the Supervisely web instance, but also provides convenience helpers for the common tasks that we found useful in our own work of developing Supervisely plugins, such as neural network architectures and custom dataset imports.

Key features:

  • Read, modify and write Supervisely projects on disk.
  • Work with labeling data: geometric objects and tags.
  • Common functionality for developing Supervisely plugins, so that you only need to focus on the core of your custom logic, and not on low level interactions with the platform.

Installation:

pip install supervisely

or

git clone https://github.com/supervisely/supervisely.git && \
pip install -e ./supervisely

or

python -m pip install git+https://github.com/supervisely/supervisely.git

We release updates quite often, so use following command if you would like to upgrade you current supervisely package:

pip install supervisely --upgrade

The only prerequisites are Python >= 3.6 and pip.

Tip: opencv-python may require apt-get install libgtk2.0-dev Or use pre-built Docker image which can be found on Docker Hub:

docker pull supervisely/base-py

The corresponding Dockerfile can be found in base_images directory.

Data Transformation Language

Data Transformation Language allows to automate complicated pipelines of data transformation. Different actions determined by DTL layers may be applied to images and annotations. In details it is described here.

Resources

  • Supervise.ly - Website
  • Medium - Our technical blog. Regular updates on how to use state of the art models and solve practical data science problems with Supervisely.
  • Tutorials and Cookbooks in this repository.