论文标题
长尾视觉识别的平衡激活
Balanced Activation for Long-tailed Visual Recognition
论文作者
论文摘要
深层分类器在视觉识别方面取得了巨大的成功。但是，现实世界中的数据长期以来是大自然的，导致培训和测试分布之间的不匹配。在本报告中，我们引入了平衡的激活（平衡的软磁体和平衡的Sigmoid），这是Sigmoid和SoftMax激活功能的优雅无偏，简单的扩展，以适应对象检测中训练和测试之间的标签分布变化。我们得出了多类软性超级回归的概括结合，并显示我们的损失最大程度地减少了界限。在我们的实验中，我们证明了平衡激活通常在LVIS-1.0上的MAP方面可提供约3％的增益，并且在没有引入任何额外参数的情况下胜过当前最新方法。
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 report, we introduce Balanced Activation (Balanced Softmax and Balanced Sigmoid), an elegant unbiased, and simple extension of Sigmoid and Softmax activation function, to accommodate the label distribution shift between training and testing in object detection. We derive the generalization bound for multiclass Softmax regression and show our loss minimizes the bound. In our experiments, we demonstrate that Balanced Activation generally provides ~3% gain in terms of mAP on LVIS-1.0 and outperforms the current state-of-the-art methods without introducing any extra parameters.
