Balanced Meta-Softmax for Long-Tailed Visual Recognition
This work addresses the mismatch between training and testing distributions in real-world long-tailed data for visual recognition, offering an incremental improvement over existing solutions.
The paper tackles the problem of long-tailed visual recognition by addressing biased gradient estimation in Softmax under label distribution shifts, introducing Balanced Meta-Softmax which outperforms state-of-the-art methods on classification and instance segmentation tasks.
Deep classifiers have achieved great success in visual recognition. However, real-world data is long-tailed by nature, leading to the mismatch between training and testing distributions. In this paper, we show that the Softmax function, though used in most classification tasks, gives a biased gradient estimation under the long-tailed setup. This paper presents Balanced Softmax, an elegant unbiased extension of Softmax, to accommodate the label distribution shift between training and testing. Theoretically, we derive the generalization bound for multiclass Softmax regression and show our loss minimizes the bound. In addition, we introduce Balanced Meta-Softmax, applying a complementary Meta Sampler to estimate the optimal class sample rate and further improve long-tailed learning. In our experiments, we demonstrate that Balanced Meta-Softmax outperforms state-of-the-art long-tailed classification solutions on both visual recognition and instance segmentation tasks.