Figure Eight Federal’s patented, cutting-edge video annotation solution combines machine learning with annotation by skilled human annotators to create training data labels that track objects moving through space and time up to 100 times faster than human-only solutions.
In the first frame of a video, our human labelers annotate the objects of interest, as defined by user classes. Functionally, this step is like a typical image annotation workflow. What makes Figure Eight Federal’s Machine Learning Assisted Video Object Tracking (MLAVOT) solution truly powerful is what comes next:
Using a deep learning ensemble model, our MLAVOT solution predicts where all annotated objects will move in subsequent frames. Each individual label persists, even if there are dozens of instances of the same class. Instead of having to relabel the entire image from scratch on each subsequent frame, our human annotator simply inspects and corrects the annotations as necessary, dragging or resizing the persisted label to squarely fit around the annotated object.
Once an annotator has labeled each object in the first frame, an ensemble deep learning model maintains those labels onto subsequent frames. Annotators therefore need only make small corrections on subsequent frames, instead of labeling every object again, making our MLAVOT solution up to 100 times faster than human-only approaches.
Our MLAVOT tool supports bounding box, polygon, dot, and line annotation schemes, enabling a wide array of use cases. Users can track almost any moving object, such as cars, road lane lines, body parts, flying objects, and much more.
Our MLAVOT solution allows you to create a custom ontology of up to 255 classes, specific to your use cases. We also support multiple object instances in each class so you can label even complex images.