Browse Teach Curate

Missing Views as Data Augmentation in Multi-View Earth Observation Learning

A multi-view learning approach for Earth observation that treats missing views as an augmentation signal rather than a defect: it trains on enumerated combinations of present and absent views and merges available views with dynamic functions (averaging up to a Transformer) so the model can ignore missing views. This yields a single adaptive model robust to any subset of available views.

Demerzel Text

MISSING DATA AS AUGMENTATION IN THE EARTH OBSERVATION DOMAIN: A MULTI-VIEW LEARNING APPROACH A

In Earth-observation problems a model often combines several heterogeneous views (optical and radar satellite image time series, weather, topography, metadata), but some views are frequently missing …