Recognizing cultural events in images: A study of image categorization models

The goal of this work is to study recognition of cultural events represented in still images. We pose cultural event recognition as an image categorization problem, and we study the performance of several state-of-the-art image categorization approaches, including Spatial Pyramid Matching and Regularized Max Pooling. We consider SIFT and color features as well as the recently proposed CNN features. Experiments on the ChaLearn dataset of 50 cultural events, we find that Regularized Max Pooling with CNN, SIFT, and Color features achieves the best performance.