SDK Developer Guide

Find the latest support doc here!

Rendered.ai’s SDK: anatools

anatools is an SDK for connecting to the Rendered.ai Platform. With anatools you can generate and access synthetic datasets, and much more!

>>> import anatools
>>> ana = anatools.client()
'Enter your credentials for the Rendered.ai Platform.'
'email:' example@rendered.ai
'password:' ***************
>>> channels = ana.get_channels()
>>> graphs = ana.get_staged_graphs()
>>> datasets = ana.get_datasets()


Install the anatools Package

(Optional) Create a new Conda Environment

  1. Install conda for your operating system: https://www.anaconda.com/products/individual.

  2. Create a new conda environment and activate it.

  3. Install anatools from the Python Package Index.

$ conda create -n renderedai python=3.7
$ conda activate renderedai

Install AnaTools to the Python Environment

  1. Install AnaTools from the Python Package Index.

$ pip install anatools

Dependencies

The anatools package requires python 3.6 or higher and has dependencies on the following packages:

Package

Description

docker

A python library for the Docker Engine API.

numpy

A python library used for array-based processing.

pillow

A fork of the Python Image Library.

pyyaml

A python YAML parser and emitter.

requests

A simple HTTP python library.

If you have any questions or comments, contact Rendered.AI at info@rendered.ai.


Quickstart Guide

What is the Rendered.ai Platform?

The Rendered.ai Platform is a synthetic dataset generation tool where graphs describe what and how synthetic datasets are generated.

Terms

Definitions

workspace

A workspace is a collection of data used for a particular use-case, for example workspaces can be used to organize data for different projects.

dataset

A dataset is a collection of data, for many use-cases these are images with text-based annotation files.

graph

A graph is defined by nodes and links, it describes the what and the how a dataset is generated.

node

A node can be described as an executable block of code, it has inputs and runs some algorithm to generate outputs.

link

A link is used to transfer data from the output of one node, to the input of other nodes.

channel

A channel is a collection of nodes, it is used to limit the scope of what is possible to generate in a dataset (like content from a tv channel).

How do you use the SDK?

The Rendered.ai Platform creates synthetic datasets by processing a graph, so we will need to create the client to connect to the Platform API, create a graph, then create a dataset.

  1. Execute the python command line, create a client and login to Rendered.ai. In this example we are instantiating a client with no workspace or environment variables, so it is setting our default workspace. To access the tool, you will need to use your email and password for https://deckard.rendered.ai.

    >>> import anatools
    >>> ana = anatools.client()
    'Enter your credentials for the Rendered.ai Platform.'
    'email:' example@rendered.ai
    'password:' ***************
    
  2. Create a graph file called graph.yml with the code below. We are defining a simplistic graph for this example with multiple children’s toys dropped into a container. While YAML files are used in channel development and for this example, the Platform SDK and API only support JSON. Ensure that the YAML file is valid in order for the SDK to convert YAML to JSON for you. Otherwise, provide a graph in JSON format.

    version: 2
    nodes:
    
    Rubik's Cube:
     nodeClass: "Rubik's Cube"
    
    Mix Cube:
     nodeClass: Mix Cube
    
    Bubbles:
     nodeClass: Bubbles
    
    Yoyo:
     nodeClass: Yo-yo
    
    Skateboard:
     nodeClass: Skateboard
    
    MouldingClay:
      nodeClass: Playdough
    
    ColorToys:
     nodeClass: ColorVariation
     values: {Color: "<random>"}
     links:
       Generators:
         - {sourceNode: Bubbles, outputPort: Bubbles Bottle Generator}
         - {sourceNode: Yoyo, outputPort: Yoyo Generator}
         - {sourceNode: MouldingClay, outputPort: Play Dough Generator}
         - {sourceNode: Skateboard, outputPort: Skateboard Generator}
    
    ObjectPlacement:
     nodeClass: RandomPlacement
     values: {Number of Objects: 20}
     links:
       Object Generators:
       - {sourceNode: ColorToys, outputPort: Generator}
       - {sourceNode: "Rubik's Cube", outputPort: "Rubik's Cube Generator"}
       - {sourceNode: Mix Cube, outputPort: Mixed Cube Generator}
    
    Container:
     nodeClass: Container
     values: {Container Type: "Light Wooden Box"}
    
    Floor:
     nodeClass: Floor
     values: {Floor Type: "Granite"}
    
    DropObjects:
     nodeClass: DropObjectsNode
     links:
       Objects:
         - {sourceNode: ObjectPlacement, outputPort: Objects}
       Container Generator:
         - {sourceNode: Container, outputPort: Container Generator}
       Floor Generator:
         - {sourceNode: Floor, outputPort: Floor Generator}
    
    Render:
     nodeClass: RenderNode
     links:
       Objects of Interest:
       - {sourceNode: DropObjects, outputPort: Objects of Interest}
    
  3. Create a graph using the client. To create a new graph, we load the graph defined above into a python dictionary using the yaml python package. Then we create a graph using the client. This graph is being named testgraph and is using the example channel. We will first find the channelId matching to the example channel and use that in the create_staged_graph call. The client will return a graphId so we can reference this graph later.

    >>> import yaml
    >>> with open('graph.yml') as graphfile:
    >>>     graph = yaml.safe_load(graphfile)
    >>> channels = ana.get_channels()
    >>> channelId = list(filter(lambda channel: channel['name'] == 'example', channels))[0]['channelId']
    >>> graphId = ana.create_staged_graph(name='testgraph', channelId=channelId, graph=graph)
    >>> print(graphId)
    '010f9362-daa8-4c10-a3e8-1e81e0f2e4f4'
    
  4. Create a dataset using the client. Using the graphId, we can create a new job to generate a dataset. The job takes some time to run.

The client will return a datasetId that can be used for reference later. You can use this datasetId to check the job status and, once the job is complete, download the dataset. You have now generated Synthetic Data!

>>> datasetId = ana.create_dataset(name='testdataset',graphId=graphId,interpretations='10',priority='1',seed='1',description='A simple dataset with cubes in a container.')
>>> datasetId
'ce66e81c-23a6-11eb-adc1-0242ac120002'

Indices and tables