For supervised machine learning, one needs lots of annotated, or labelled data, meaning an image with object labels and the precise pixels occupied by those objects. Large, diverse, high-fidelity training datasets are essential to autonomous vehicles using deep learning for scene understanding.
The safety of autonomous agents is predicated on their flawless environment sensing and perception, which require verification and certification using their sensor-specific data.
However, the scene diversity and variety of edge cases achievable with physically driveable mileage are insufficient for fully autonomous driving certification.
We are developing a breakthrough technology, which we call STRADA, that takes the best of the synthetic and hand-labelled data approaches.
It involves automatic augmentation of real-world data to generate high-fidelity, high-diversity mixed-reality data that does not require the expensive human labelling or post-processing.
By augmenting real-world imagery in multiple sensor modalities, we can achieve greater realism and diversity than synthetic data from computer-generated proxy worlds.
Our image augmentation technology will generate annotated datasets of real-world imagery in different sensing modalities (visible, night vision / IR, LiDAR, radar, or a custom imager) with automatic instance-level object segmentation in complex multi-object scenes at high resolution.
Our sensor-specific data will be offered to vehicle OEMs, Tier 1 suppliers and automotive technology providers for training deep neural networks in scene understanding.
We will also build geo-targeted datasets to verify and certify the functional safety of environmental perception for a complete sensor set on an autonomous vehicle.