The Data Operations Platform for Machine Perception
We are committed to simplifying machine learning data operations management for perception-heavy applications such as automated driving, robotics, remote sensing, and surveillance.
Our flagship product, SCENEBOX, empowers perception teams to build better training datasets faster, enables them to focus on core ML tasks rather than data operations, and significantly reduces ML product time to market.
Edge to core platform
SCENEBOX characterizes data at the edge which allows you to transfer only the useful perception data to the core platform.
What does this mean?
You can centrally access your distributed data regardless of where it resides (on cloud, on-prem, or at the edge).
Pre-label at scale
You can automatically label your perception data using our library of enrichments (such as object detection, weather condition, etc.), or by bringing your own model (BYOM) or even your labels (BYOL).
Perception data is composed of multiple streams of data from various sources. SCENEBOX's event engine uses event time for correlating data across an arbitrarily large number of sources and allows for queries such as:
"Give me all LiDAR scenes where a car and a pedestrian were observed from a side camera and where the vehicle has a speed larger than 50 km/h".
Every piece of data in SCENEBOX is indexed and is searchable across all of its dimensions using a powerful search interface.
SCENEBOX can manage standard perception data such as RGB images, videos, ROS, multi-band satellite imagery as well as data from novel sensors such as RGBD, new generations of LiDAR, and more.
Any 3rd party annotation solution
Through our annotation hub you have access to best in class annotation providers such as Scale, LabelBox, Deepen, etc. With one click your curated datasets are sent to a provider of your choice, and the annotated data is ingested automatically into SCENEBOX.
Stringent data requirements such as data privacy, residency, or volume often dictate the deployment strategy. SCENEBOX's cloud-agnostic implementation allows you to either use Caliber's multi-tenant cloud to support data operations or deploy SCENEBOX to your private servers. Your servers can be on any cloud (AWS, GCP, or Azure) or in an on-prem air-gapped environment.