aiData
aiData is a streamlined, automated data pipeline designed to optimize the development of automated driving technology. By automating key stages in the Machine Learning Operations (MLOps) workflow – from data collection and preparation to curation, annotation, and validation – aiData ensures a smooth transition of data between data scientists and developers.
This efficiency boost of the automotive MLOps workflow allows for faster and more effective deployment of ADAS/AD models into production.
aiData Versioning System
Commonly, data preparation takes a significant amount of time in the MLOps workflow since cleaning up and curating the training datasets is highly resource intensive.
aiData Versioning System provides complete transparency and traceability over the entire data flow, enables the curation of datasets with latest AI-based technologies, as text, image and scenario-based searching.
- Keep track of the whole data journey from recording through annotation then adding to a training or validation dataset
- Curate datasets for a variety of use-cases, based on enriched metadata (e.g. weather, cartography), scene contents and efficient SQL-based queries
- Enjoy full traceability during the product cycle
- Manage data recorded by external loggers or the aiData Recorder, all utilizing the data enrichment of the Versioning System, such as auto-tagging weather, cartography and other content parameters
- Deploy on premise for the highest security or in the cloud for easy collaboration within global teams
aiData Recorder
Recording diverse scenarios with precise sensor calibration and synchronization is critical to produce high-quality data for automated driving development, testing and validation.
The aiData Recorder is an adaptable smart data collection software to ensure that the recorded data is of the highest quality:
- Adaptable to customer specific sensor configurations
- Reference implementation on various vehicle models with various sensor setups
- Complete offline and on-the-fly calibration for multimodal sensor setups
- Precise time synchronization of the sensor recordings
- Server based backend and UI for recording management
- Recorded data is automatically uploaded to the Versioning System where it can be curated and analyzed, then sent for manual or automatic annotation
aiData Auto Annotator
Data annotation is a traditionally highly manual and resource intensive process due to the vast amounts of data involved. With the aiData Annotator annotation can be done automatically, within hours of the original recording.
- Multi sensor auto annotation for dynamic and static objects, utilizing AI algorithms and a GPU cluster
- Real-world scenario extraction converts raw sensor data into virtual scenarios for closed-loop simulation
- Simultaneous annotation of all sensors, including Lidar point cloud and camera images, with a consistent 4D (space + time) environment model
- 100% precision for static object detection and 90%+ precision for dynamic object detection
- Built-in quality control measures, including automated and manual quality control
aiData Metrics
Data validation can be a complex process to ensure that real-world conditions are reflected correctly in the datasets. With aiData Metrics measure development progress against requirements and dive into real-time insights and data gap analysis.
- Spot data gaps and determine which data is useful easily
- Comprehensive tool for evaluation of neural network (NN) algorithms and detection software, most popular use-cases include:
- Environment detection benchmarks
- Object tracking benchmarks to evaluate perception algorithms
- Built-in visualization for quick assessment in flexible input / output formats
