Computational Recognition of Facial Expressions in Sculpture

Our Contributions

• We utilized deep learning-based methods to analyze facial expressions in sculpture faces.
• We collected images from different museums worldwide and introduced a publicly available automatically labeled dataset for the research community.
• The existing FER methods, pre-trained on human facial expressions, generate soft and multi-label classes for the collected dataset.
• We designed deep convolutional neural network (DCNN) based models, using different strategies to classify the sculpture’s expressions into multiple classes.
• We also generated coarse localization maps to visualize the important image regions that contributed to the predictions

Paper Available Here