GestaltMML
A multimodal deep learning model for facial phenotype analysis of genetic syndromes. It integrates facial images with other modalities and, in synthetic-data oversampling experiments, shows more sustained gains than image-based models.
Timeline (last 90 days)
The study compared two model architectures: the image-based GestaltMatcher-Arc and the multimodal GestaltMML.
Synthetic oversampling improved the mean recognition accuracy under both training settings.
The accuracy gain for GestaltMatcher-Arc was limited, while GestaltMML benefited consistently at every oversampling level tested.
All Facts (3)
The study compared two model architectures: the image-based GestaltMatcher-Arc and the multimodal GestaltMML.
50%UnverifiedSynthetic oversampling improved the mean recognition accuracy under both training settings.
50%UnverifiedThe accuracy gain for GestaltMatcher-Arc was limited, while GestaltMML benefited consistently at every oversampling level tested.
50%