Visual Sentiment Prediction Using Few-shot Learning via Distribution Relations of Visual Features
Yingrui Ye, Yuya Moroto, Keisuke Maeda, Takahiro Ogawa, Miki Haseyama · 2021 IEEE 10th Global Conference on Consumer Electronics (GCCE) · 2021
A visual sentiment prediction method using few-shot learning via distribution relations of visual features is proposed in this paper. To train a model that can predict the sentiments of images with a small amount of labeled data, the proposed method focuses on distribution relations of feature embeddings of images. The relations are helpful for visual sentiment prediction since distances between different sentiments are not equal according to Mikels’ wheel. The proposed method incorporates distribution propagation in a graph network to take advantage of the distribution relations. The experiment shows that the prediction accuracy of the proposed method achieves an improvement compared to comparative methods. The main contribution of this paper is the introduction of a new method using few-shot learning via distribution relations of visual features for visual sentiment prediction.