Caffe：你如何为测试层打印加权损失？

Question

我当前的 Caffe 输出如下所示：

Iteration 1000, Testing net (#0)
Test net output #0: accuracy_1 = 0.337018
Test net output #1: accuracy_2 = 0.3397
Test net output #2: accuracy_3 = 0.360761
Test net output #3: loss_1 = 2.08132 (* 1 = 2.08132 loss)
Test net output #4: loss_2 = 2.03755 (* 1 = 2.03755 loss)
Test net output #5: loss_3 = 1.91984 (* 1 = 1.91984 loss)

Iteration 1000, loss = 3.87841
Train net output #0: loss_1 = 1.26657 (* 1 = 1.26657 loss)
Train net output #1: loss_2 = 1.40096 (* 1 = 1.40096 loss)
Train net output #2: loss_3 = 1.21088 (* 1 = 1.21088 loss)

训练迭代会打印出正确的加权损失（又名“loss = 3.87841”），而测试迭代只会显示“Testing net (#0)”。如何让测试迭代也打印出正确的加权损失？谢谢！

Answer 1

我不认为这是你的加权损失；但它是在迭代中平滑的训练的平均损失。

您可能需要在此处检查 average_loss 字段： https://github.com/BVLC/caffe/blob/master/src/caffe/solver.cpp#L190-L221

分享您的损失和准确率层以获得更详细的答案。