Furniture Style Compatibility Estimation by Multi-Branch Deep Siamese Network

As demands for understanding visual style among interior scenes increase, estimating compatibility is becoming challenging. In particular, furniture styles are difficult to define due their various elements, such as color and shape. a result, an ambiguous concept. To reduce ambiguity, Siamese networks have frequently been used estimate by adding features that represent the style. However, it still accurately furniture’s style, even when using alternate associated with images. this paper, we propose new model can learn from several images simultaneously. Specifically, one-to-many ratio input method maintain high performance inputs ambiguous. We also metric evaluating networks. The conventional metric, area under ROC curve (AUC), does not reveal actual distance between styles. Therefore, proposed quantitatively evaluates embedding of each image. Experiments show improved AUC 0.672 0.721 outperformed in terms metric.