Ana

Ana is the code framework for running jobs in Rendered.ai, a Platform as a Service (PaaS) for generating synthetic data that enables data scientists and engineers to overcome the costs and challenges of acquiring and using real data for training machine learning and artificial intelligence systems. Rendered.ai has open sourced Ana becuase it is the basis for creating custom channels to generate specific types of synthetic datasets.

Channels contain configuration and customization for the sensor models, platform models, and scene content that are used to generate a wide range of simulated sensor outputs that are grouped into a synthetic dataset. Developers use Ana to build and test channels, deploy channels to the Rendered.ai engine, and then use Rendered.ai either through the web interface or through automated API calls to generate datasets that can be used for AI/ML training, validation, and testing.

For more information on the Rendered.ai platform and why synthetic data is important for AI, visit the Rendered.ai support pages.

The source code and files in this repository are copyright 2019-2022 DADoES, Inc. and licensed under the Apache 2.0 license which is located at the root level of this repository: LICENSE.

Setup

Setting Up the Development Environment: Rendered.ai provides Development Docker Images that can simplify setting up your environment for channel development and can accelerate the channel deployment process. A development container (VSCode extension) saves you the trouble of building the channel every time you want to test, and ensures the code you are working with is the code that is in the docker when it is registered.

Running Ana From the Command Line

For convenience during developing a channel, Ana can be run locally from the command line. To run a graph in a channel a single time use the following command line syntax:

blender --background --python ana.py -- --channel <channel-name> --graph <graph-file> --loglevel <log-level> \
 --data <data-dir> --output <output-dir>

The arguments are

• <channel-name>: the name of the channel
• <graph-file>: the name of the graph file in ./channels/<channel-name>/graphs/ - supported formats are JSON and YAML
• <data-dir>: the relative location of object model assets
• <output-dir>: the relative location to write the result of the run
• <log-level>: the Python logging levels

Running the Example Channel

All accounts at Rendered.ai come with an example channel that shows the Ana tool and common libraries. Developers can inspect the code example channel code, test it, and see how the channel can be used when deployed to Rendered.ai. For example, the graph in this repository produces images of objects dropped into a container that sits on a floor.

ana/ana $ blender --background --python ana.py -- --channel example --graph example_test --loglevel INFO

And here is how to run the Example Channel Container

$ docker run --gpus all -v $ANA_ROOT/ana/datasets:/ana/datasets example blender --background --python ana/ana.py -- \
 --channel example --graph ana/channels/example/graphs/example_test.yml --data /data --output /ana/datasets

After running a graph in the channel, an output folder will be created with the following directories:

output/
    |- annotations/
    |- images/
    |- masks/
    |- metadata/

The image below is an example of what you'll find in the images/ directory.

The available graphs for the channel are located in <ana-root>/ana/channels/example/graphs.

graph	description
example_test	A simple test graph that drops some toys into a wooden box.

Additional Resources

Setting up your Local Environment
Ana Software Architecture
Creating a Channel
Deploying a Channel
Adding Content to a Channel
Managing Content with Volumes