Caffe-Python接口常用API参考

官网也有提供demo

http://nbviewer.jupyter.org/github/BVLC/caffe/blob/master/examples/net_surgery.ipynb

本文整理了pycaffe中常用的API

Packages导入

123	import caffefrom caffe import layers as Lfrom caffe import params as P

Layers定义

Data层定义

lmdb/leveldb Data层定义

12345678910

L.Data( source=lmdb, backend=P.Data.LMDB, batch_size=batch_size, ntop=2, transform_param=dict( crop_size=227, mean_value=[104, 117, 123], mirror=True ) )

HDF5 Data层定义

123456789

L.HDF5Data( hdf5_data_param={ 'source': './training_data_paths.txt', 'batch_size': 64 }, include={ 'phase': caffe.TRAIN } )

ImageData Data层定义

适用于txt文件一行记录一张图片的数据源

12345678

L.ImageData( source=list_path, batch_size=batch_size, new_width=48, new_height=48, ntop=2, ransform_param=dict(crop_size=40,mirror=True) )

Convloution层定义

12345678

L.Convolution( bottom, kernel_size=ks, stride=stride, num_output=nout, pad=pad, group=group )

LRN层定义

123456

L.LRN( bottom, local_size=5, alpha=1e-4, beta=0.75 )

Activation层定义

ReLU层定义

1234	L.ReLU( bottom, in_place=True )

Pooling层定义

123456

L.Pooling( bottom, pool=P.Pooling.MAX, kernel_size=ks, stride=stride )

FullConnect层定义

1234	L.InnerProduct( bottom, num_output=nout )

Dropout层定义

1234	L.Dropout( bottom, in_place=True )

Loss层定义

1234	L.SoftmaxWithLoss( bottom, label )

Accuracy层定义

1234	L.Accuracy( bottom, label )

转换为proto文本

1234	caffe.to_proto( loss, acc #训练阶段可以删去Accuracy层 )

Solver定义

123456789101112131415161718192021222324252627

from caffe.proto import caffe_pb2s = caffe_pb2.SolverParameter()path='/home/xxx/data/'solver_file=path+'solver.prototxt' #solver文件保存位置s.train_net = path+'train.prototxt' # 训练配置文件s.test_net.append(path+'val.prototxt') # 测试配置文件s.test_interval = 782 # 测试间隔s.test_iter.append(313) # 测试迭代次数s.max_iter = 78200 # 最大迭代次数s.base_lr = 0.001 # 基础学习率s.momentum = 0.9 # momentum系数s.weight_decay = 5e-4 # 权值衰减系数s.lr_policy = 'step' # 学习率衰减方法s.stepsize=26067 # 此值仅对step方法有效s.gamma = 0.1 # 学习率衰减指数s.display = 782 # 屏幕日志显示间隔s.snapshot = 7820s.snapshot_prefix = 'shapshot's.type = “SGD” # 优化算法s.solver_mode = caffe_pb2.SolverParameter.GPUwith open(solver_file, 'w') as f: f.write(str(s))

Model训练

1234567891011121314151617181920212223

# 训练设置# 使用GPUcaffe.set_device(gpu_id) # 若不设置,默认为0caffe.set_mode_gpu()# 使用CPUcaffe.set_mode_cpu()# 加载Solver，有两种常用方法# 1. 无论模型中Slover类型是什么统一设置为SGDsolver = caffe.SGDSolver('/home/xxx/data/solver.prototxt') # 2. 根据solver的prototxt中solver_type读取，默认为SGDsolver = caffe.get_solver('/home/xxx/data/solver.prototxt')# 训练模型# 1.1 前向传播solver.net.forward() # train netsolver.test_nets[0].forward() # test net (there can be more than one)# 1.2 反向传播,计算梯度solver.net.backward()# 2. 进行一次前向传播一次反向传播并根据梯度更新参数solver.step(1)# 3. 根据solver文件中设置进行完整model训练solver.solve()

如果想在训练过程中保存模型参数，调用

1	solver.net.save('mymodel.caffemodel')

分类图片

加载Model数据

12345

net = caffe.Net( deploy_prototxt_path, # 用于分类的网络定义文件路径 caffe_model_path, # 训练好模型路径 caffe.TEST # 设置为测试阶段 )

中值文件转换

12345678910111213

#　编写一个函数，将二进制的均值转换为python的均值def convert_mean(binMean,npyMean): blob = caffe.proto.caffe_pb2.BlobProto() bin_mean = open(binMean, 'rb' ).read() blob.ParseFromString(bin_mean) arr = np.array( caffe.io.blobproto_to_array(blob) ) npy_mean = arr[0] np.save(npyMean, npy_mean )# 调用函数转换均值binMean='examples/cifar10/mean.binaryproto'npyMean='examples/cifar10/mean.npy'convert_mean(binMean,npyMean)

图片预处理

12345678910111213141516

# 设定图片的shape格式为网络data层格式transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape})# 改变维度的顺序，由原始图片维度(width, height, channel)变为(channel, width, height)transformer.set_transpose('data', (2,0,1)) # 减去均值,注意要先将binaryproto格式均值文件转换为npy格式[此步根据训练model时设置可选]transformer.set_mean('data', np.load(mean_file_path).mean(1).mean(1))# 缩放到[0，255]之间transformer.set_raw_scale('data', 255)# 交换通道，将图片由RGB变为BGRtransformer.set_channel_swap('data', (2,1,0))# 加载图片im=caffe.io.load_image(img)# 执行上面设置的图片预处理操作，并将图片载入到blob中net.blobs['data'].data[...] = transformer.preprocess('data',im)

执行测试

123456789101112

#执行测试out = net.forward()labels = np.loadtxt(labels_filename, str, delimiter='\t') #读取类别名称文件prob= net.blobs['Softmax1'].data[0].flatten() #取出最后一层（Softmax）属于某个类别的概率值，并打印print proborder=prob.argsort()[0] #将概率值排序，取出最大值所在的序号 print 'the class is:',labels[order] #将该序号转换成对应的类别名称，并打印# 取出前五个较大值所在的序号top_inds = prob.argsort()[::-1][:5]print 'probabilities and labels:' zip(prob[top_inds], labels[top_inds])

各层信息显示

1234567

# params显示：layer名，w，bfor layer_name, param in net.params.items(): print layer_name + '\t' + str(param[0].data.shape), str(param[1].data.shape)# blob显示：layer名，输出的blob维度for layer_name, blob in net.blobs.items(): print layer_name + '\t' + str(blob.data.shape)

自定义函数:参数/卷积结果可视化

123456789101112131415161718192021222324252627282930313233343536

import numpy as npimport matplotlib.pyplot as pltimport matplotlib.image as mpimgimport caffe%matplotlib inlineplt.rcParams['figure.figsize'] = (8, 8)plt.rcParams['image.interpolation'] = 'nearest'plt.rcParams['image.cmap'] = 'gray'def show_data(data, padsize=1, padval=0):"""Take an array of shape (n, height, width) or (n, height, width, 3) and visualize each (height, width) thing in a grid of size approx. sqrt(n) by sqrt(n)""" # data归一化 data -= data.min() data /= data.max() # 根据data中图片数量data.shape[0]，计算最后输出时每行每列图片数n n = int(np.ceil(np.sqrt(data.shape[0]))) # padding = ((图片个数维度的padding),(图片高的padding), (图片宽的padding), ....) padding = ((0, n ** 2 - data.shape[0]), (0, padsize), (0, padsize)) + ((0, 0),) * (data.ndim - 3) data = np.pad(data, padding, mode='constant', constant_values=(padval, padval)) # 先将padding后的data分成n*n张图像 data = data.reshape((n, n) + data.shape[1:]).transpose((0, 2, 1, 3) + tuple(range(4, data.ndim + 1))) # 再将（n, W, n, H）变换成(n*w, n*H) data = data.reshape((n * data.shape[1], n * data.shape[3]) + data.shape[4:]) plt.figure() plt.imshow(data,cmap='gray') plt.axis('off')# 示例：显示第一个卷积层的输出数据和权值（filter）print net.blobs['conv1'].data[0].shapeshow_data(net.blobs['conv1'].data[0])print net.params['conv1'][0].data.shapeshow_data(net.params['conv1'][0].data.reshape(32*3,5,5))

自定义：训练过程Loss&Accuracy可视化

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768

import matplotlib.pyplot as plt import caffe caffe.set_device(0) caffe.set_mode_gpu() # 使用SGDSolver，即随机梯度下降算法 solver = caffe.SGDSolver('/home/xxx/mnist/solver.prototxt') # 等价于solver文件中的max_iter，即最大解算次数 niter = 10000 # 每隔100次收集一次loss数据 display= 100 # 每次测试进行100次解算 test_iter = 100# 每500次训练进行一次测试test_interval =500 #初始化 train_loss = zeros(ceil(niter * 1.0 / display)) test_loss = zeros(ceil(niter * 1.0 / test_interval)) test_acc = zeros(ceil(niter * 1.0 / test_interval)) # 辅助变量 _train_loss = 0; _test_loss = 0; _accuracy = 0 # 进行解算 for it in range(niter): # 进行一次解算 solver.step(1) # 统计train loss _train_loss += solver.net.blobs['SoftmaxWithLoss1'].data if it % display == 0: # 计算平均train loss train_loss[it // display] = _train_loss / display _train_loss = 0 if it % test_interval == 0: for test_it in range(test_iter): # 进行一次测试 solver.test_nets[0].forward() # 计算test loss _test_loss += solver.test_nets[0].blobs['SoftmaxWithLoss1'].data # 计算test accuracy _accuracy += solver.test_nets[0].blobs['Accuracy1'].data # 计算平均test loss test_loss[it / test_interval] = _test_loss / test_iter # 计算平均test accuracy test_acc[it / test_interval] = _accuracy / test_iter _test_loss = 0 _accuracy = 0 # 绘制train loss、test loss和accuracy曲线 print '\nplot the train loss and test accuracy\n' _, ax1 = plt.subplots() ax2 = ax1.twinx() # train loss -> 绿色 ax1.plot(display * arange(len(train_loss)), train_loss, 'g') # test loss -> 黄色 ax1.plot(test_interval * arange(len(test_loss)), test_loss, 'y') # test accuracy -> 红色 ax2.plot(test_interval * arange(len(test_acc)), test_acc, 'r') ax1.set_xlabel('iteration') ax1.set_ylabel('loss') ax2.set_ylabel('accuracy') plt.show()