Interactive Visual Study of Multiple Attributes Learning Model of X-Ray Scattering Images

Existing interactive visualization tools for deep learning are mostly applied to the training, debugging, and refinement of neural network models working on natural images. However, visual analytics lacking specific application x-ray image classification with multiple structural attributes. In this paper, we present an system domain scientists visually study attributes scattering It allows interactively explore important type scientific images in embedded spaces that defined model prediction output, actual labels, discovered feature space networks. Users allowed flexibly select instance images, their clusters, compare them regarding specified representation The exploration is guided by manifestation performance related mutual relationships among attributes, which often affect accuracy effectiveness. thus supports improve training dataset model, find questionable identify outlier or spurious data clusters. Case studies feedback demonstrate its functionalities usefulness.