The Object Detection activity developed domain-specific solutions as well as general methods addressing the need to detect data items of interest, or extract or estimate certain properties.
For example, ML algorithms traditionally require large amounts of labelled data to learn from, and one of the categories, Small Data, concerns itself with inherently reducing the reliance on labelled data.
In contrast, the Synthetic Data category addresses a related problem by artificially producing more training data when the circumstances allow for modelling the data of interest appropriately.
A third category, Multimodal, leverages the fact that sometimes the information we need to detect/extract is spread across different complementary data streams.
In the case of the Temporal category, the information needed to identify what we are interested in is dispersed sequentially in the time domain and requires yet another approach.
The Object Detection Activity worked with CSIRO Business Units to identify projects aligned with these categories and to leverage synergies. The data-agnostic ML Core of the project was intended to ensure that learnings in one part of the activity are generalised and reusable in other parts.