Multi-Expert Cooperative and Feature Optimization Network for Long-Tailed Data

In long-tailed visual classification tasks, the imbalance in category distribution hinders models from adequately learning the features of tail classes. To tackle this challenge, we propose a Multi-Expert Cooperative and Feature Optimization Network. This network aims to adopt collaborative learning among multiple experts while enriching the feature information of tail classes through dual sampling. It also employs feature mapping techniques to contract the enriched feature clusters, thereby enhancing the model's learning capacity for these features. Analysis of the experimental data shows that the suggested method boosts the accuracy and reliability of classification on several long-tailed datasets. This approach provides an effective solution for improving tail class recognition and offers a new perspective on addressing the problem of long-tailed.