Learning to Aggregate and Refine Noisy Labels for Visual Sentiment Analysis

Visual sentiment analysis has received increasing attention in recent years. However, the quality of the dataset is a concern because the sentiment labels are crowd-sourcing, subjective, and prone to mistakes. This poses a severe threat to the data-driven models including the deep neural networks which would generalize poorly on the testing cases if they are trained to over-fit the samples with noisy sentiment labels. Inspired by the recent progress on learning with noisy labels, we propose a robust learning method to perform robust visual sentiment analysis. Our method relies on an external memory to aggregate and filter noisy labels during training and thus can prevent the model from overfitting the noisy cases. The memory is composed of the prototypes with corresponding labels, both of which can be updated online. We establish a benchmark for visual sentiment analysis with label noise using publicly available datasets. The experiment results of the proposed benchmark settings comprehensively show the effectiveness of our method.