1 评估

The classification performance is evaluated using two measures: the top-1 and top-5 error. 分类性能使用两个指标进行评估: top-1 错误和 top-5 错误。

The former is a multi-class classification error, the proportion of incorrectly classified images; the latter is the main evaluation criterion used in ILSVRC, and is computed as the proportion of images such that the ground-truth category is outside the top-5 prediced categories. 前者是多类分类错误, 即错误分类图像的比例; 后者是 ILSVRC 中使用的主要评估标准, 计算方法是将真实类别不在预测的前 5 个类别中的图像比例。

We compare our results with the state of the art in Table 7. 我们在表 7 中将我们的结果与现有技术进行了比较。

Our architecture achieves the best result, outperforming a single GoogLeNet by 0.9%. 我们的架构取得了最佳结果, 比单个 GoogLeNet 高出 0.9%。

The results are promising. 该结果颇具前景

2 模型描述

The main difference is that we replace the logistic regression objective with a Euclidean loss, which penalises the deviation of the predicted bounding box parameters from the ground-truth. 主要区别在于我们用欧几里得损失代替了逻辑回归目标, 该损失惩罚预测边界框参数与真实值的偏差。

The last fully-connected layer was initialised randomly and trained from scratch. 最后一个全连接层随机初始化并从头开始训练。