Automatic morphological analysis of fossils

We produced the first large image dataset of fusulinids (2400 images) with the aid of deep learning image segmentation technique (BlendMask)

We developed a combined method of deep learning and multi-view augmentation for efficient fusulinid image identification reaching >90% accuracy on generic level

We conducted advanced morphometric analysis for fusulinid taxonomy and phylogeny

We are developing multi-modal deep learning frameworks to help AI explain how they “think” when classifying fossils.