虽然我不确定如果你不定义所有这四个方法,Caffe 是否会输出错误,但你肯定需要 Setup 和 Forward:
-
设置:正是您所说的。例如,在我的准确性层中,我通常为我的整个测试集和每个样本的 softmax 概率保存一些指标(真假阳性/阴性,f 分数),以防我想组合/融合不同的网络/方法之后。这是我打开文件的地方,我将在其中写入这些信息;
-
Forward:您将在此处计算准确度本身,将预测与批次中每个样本的标签进行比较。通常这一层将有两个输入,标签(可能由数据/输入层提供的基本事实)和一个输出每个类别的批次中每个样本的预测/分数/概率的层(我通常使用 SoftMax层);
-
重塑和向后:不用担心这些。您无需担心向后传球,也无需重塑您的 blob。
这里是一个准确层的例子:
# Remark: This class is designed for a binary problem with classes '0' and '1'
# Saving this file as accuracyLayer.py
import caffe
TRAIN = 0
TEST = 1
class Accuracy_Layer(caffe.Layer):
#Setup method
def setup(self, bottom, top):
#We want two bottom blobs, the labels and the predictions
if len(bottom) != 2:
raise Exception("Wrong number of bottom blobs (prediction and label)")
#Initialize some attributes
self.correctPredictions = 0.0
self.totalImgs = 0
#Forward method
def forward(self, bottom, top):
#The order of these depends on the prototxt definition
predictions = bottom[0].data
labels = bottom[1].data
self.totalImgs += len(labels)
for i in range(len(labels)): #len(labels) is equal to the batch size
pred = predictions[i] #pred is a tuple with the normalized probability
#of a sample i.r.t. two classes
lab = labels[i]
if pred[0] > pred[1]: #this means it was predicted as class 0
if lab == 0.0:
self.correctPredictions += 1.0
else: #else, predicted as class 1
if lab == 1.0:
self.correctPredictions += 1.0
acc = correctPredictions / self.totalImgs
#output data to top blob
top[0].data = acc
def reshape(self, bottom, top):
"""
We don't need to reshape or instantiate anything that is input-size sensitive
"""
pass
def backward(self, bottom, top):
"""
This layer does not back propagate
"""
pass
以及如何在 prototxt 中定义它。在这里你会告诉 Caffe 这个层只会在 TEST 阶段出现:
layer {
name: "metrics"
type: "Python"
top: "Acc"
top: "FPR"
top: "FNR"
bottom: "prediction" #let's suppose we have these two bottom blobs
bottom: "label"
python_param {
module: "accuracyLayer"
layer: "Accuracy_Layer"
}
include {
phase: TEST. #This will ensure it will only be executed in TEST phase
}
}
顺便说一句,I've written a gist 有一个更复杂的精度 python 层示例,这可能是您正在寻找的。p>