英文原文地址 https://nbviewer.jupyter.org/github/BVLC/caffe/blob/master/examples/00-classification.ipynb
任务 Instant recognition with a pre-trained model and a tour of the net interface for visualizing features and parameters layer-by-layer.
代码及部分注释如下:
import numpy as np
import matplotlib.pyplot as plt
plt.rcParams['figure.figsize'] = (10, 10) # 设置画图尺寸
plt.rcParams['image.interpolation'] = 'nearest' # 设置插值类型
plt.rcParams['image.cmap'] = 'gray' # 设置颜色类型
import sys
caffe_root = '../'
sys.path.insert(0, caffe_root+'python') # 插在sys.path的最前面
import caffe
import os
if(os.path.isfile(caffe_root+'models/bvlc_reference_caffenet/bvlc_reference_caffenet.caffemodel')):
print('CaffeNet found')
else:
print('Downloading pre-trained CaffeNet model')
os.system(
'../scripts/download_model_binary.py ../models/bvlc_reference_caffenet')
caffe.set_mode_gpu() # 设置为gpu模式
model_def = caffe_root+'models/bvlc_reference_caffenet/deploy.prototxt'
model_weights = caffe_root + \
'models/bvlc_reference_caffenet/bvlc_reference_caffenet.caffemodel'
net = caffe.Net(model_def, # 定义模型结构
model_weights, # 包含训练权重
caffe.TEST) # 使用测试模式
mu = np.load(caffe_root+'python/caffe/imagenet/ilsvrc_2012_mean.npy')
mu = mu.mean(1).mean(1) # 对像素求均值
print('mean-subtracted values:', zip('BGR', mu))
# 设定图片的shape格式(1,3,28,28)
transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape})
# 将图片channel移动到最外侧维度上,即由原始图片(28,28,3)变为(3,28,28)
transformer.set_transpose('data', (2, 0, 1))
transformer.set_mean('data', mu) # 减去每个channel的均值
transformer.set_raw_scale('data', 255) # 将变量幅值从[0,1]变为[0,255]
transformer.set_channel_swap('data', (2, 1, 0)) # 将RGB变为BGR
net.blobs['data'].reshape(50, # batch尺寸
3, # BGR图片的3 channel
227, 227) # 图片尺寸
image = caffe.io.load_image(caffe_root+'examples/images/cat.jpg') # 加载图片
transformed_image = transformer.preprocess('data', image) # 预处理
plt.imshow(image) # 展示
net.blobs['data'].data[...] = transformed_image # 将待分类图片放入模型
output = net.forward() # 网络通过前向传播进行预测
output_prob = output['prob'][0] # 获取最大的可能概率
print('predicted class is:', output_prob.argmax()) # 获取最大的可能类型
labels_file = caffe_root+'data/ilsvrc12/synset_words.txt' # 获取标签文本
if not os.path.exists(labels_file):
os.system('../data/ilsvrc12/get_ilsvrc_aux.sh')
labels = np.loadtxt(labels_file, str, delimiter='\t') # 加载标签文本
print('output label:', labels[output_prob.argmax()])
top_inds = output_prob.argsort()[::-1][:5] # 最有可能的前五种类型
print('probabilities and labels:')
print(zip(output_prob[top_inds], labels[top_inds]))