Multimodal explanations

Modal Importance Masking

Method that sequentially masks parts of each input modality to measure the model's performance drop, thus creating a spatial and temporal importance map for each data type.

← Tillbaka