杨坤的分享空间
目录:
原始引用地址: 运行caffe识别数字的模型mnist
time: 2020.5.17 23:28
mnist是一个运用神经网络识别数字的模型,可以识别数字0到9.
对于mnist(就是数字识别)例子,参考以下地址,获取数据训练相关数据:
http://caffe.berkeleyvision.org/gathered/examples/mnist.html
准备数据:
cd $CAFFE_ROOT
./data/mnist/get_mnist.sh
./examples/mnist/create_mnist.sh
caffe在纯CPU模式下,使用多核运行
https://blog.csdn.net/b876144622/article/details/80009877
2、修改caffe目录下的Makefile.config文件,将BLAS: =atlas修改为BLAS: =open
3、再编译caffe,首先make clean,清除之前的编译结果,再依次执行make all -j16, make test -j16, make runtest -j16,编译caffe。-j16是指用16个核并行编译caffe,可以大大加快编译速度。
4、编译完成后,执行训练前,需要export OPENBLAS_NUM_THREADS=4, 即使用4个核进行训练
因为编译时,选择使用cpu,使用要更改文件:
examples/mnist/lenet_solver.prototxt
把:solver_mode: GPU 改为:solver_mode: CPU
训练命令:
export PYTHONPATH=/home/xy/works/caffe/python/
time ./examples/mnist/train_lenet.sh #vm用时17m36s j1900 42m55
(注意:识别的图片,一定要是黑底, 数字用白色写)
安装画图软件gimp:
sudo apt-get install gimp
gimp 创建27*27的bmp图:
新建file -> new 选width 27 height27 单位为像素, 底色选黑
画图:选画笔,左边size ,可以选2,size太大,27*27的画布不够画的。
导出:file->export as, 格式先jpg,
在运行python/calssify前要运行安装 protobuf:
sudo pip install protobuf
在识别图片前,需要对classify进行更改:
git diff python/classify.py
diff --git a/python/classify.py b/python/classify.py
index 4544c51..446c55e 100755
--- a/python/classify.py
+++ b/python/classify.py
@@ -105,9 +105,9 @@ def main(argv):
# Make classifier.
classifier = caffe.Classifier(args.model_def, args.pretrained_model,
- image_dims=image_dims, mean=mean,
- input_scale=args.input_scale, raw_scale=args.raw_scale,
- channel_swap=channel_swap)
+ image_dims=None, mean=None,
+ input_scale=None, raw_scale=None,
+ channel_swap=None)
# Load numpy array (.npy), directory glob (*.jpg), or image file.
args.input_file = os.path.expanduser(args.input_file)
@@ -116,11 +116,11 @@ def main(argv):
inputs = np.load(args.input_file)
elif os.path.isdir(args.input_file):
print("Loading folder: %s" % args.input_file)
- inputs =[caffe.io.load_image(im_f)
+ inputs =[caffe.io.load_image(im_f, False)
for im_f in glob.glob(args.input_file + '/*.' + args.ext)]
else:
print("Loading file: %s" % args.input_file)
- inputs = [caffe.io.load_image(args.input_file)]
+ inputs = [caffe.io.load_image(args.input_file, False)]
print("Classifying %d inputs." % len(inputs))
@@ -131,6 +131,7 @@ def main(argv):
# Save
print("Saving results into %s" % args.output_file)
+ print(predictions)
np.save(args.output_file, predictions)
git diff python/caffe/classifier.py
diff --git a/python/caffe/classifier.py b/python/caffe/classifier.py
index 64d804be..65b0d881 100644
--- a/python/caffe/classifier.py
+++ b/python/caffe/classifier.py
@@ -69,18 +69,18 @@ class Classifier(caffe.Net):
for ix, in_ in enumerate(inputs):
input_[ix] = caffe.io.resize_image(in_, self.image_dims)
- if oversample:
- # Generate center, corner, and mirrored crops.
- input_ = caffe.io.oversample(input_, self.crop_dims)
- else:
- # Take center crop.
- center = np.array(self.image_dims) / 2.0
- crop = np.tile(center, (1, 2))[0] + np.concatenate([
- -self.crop_dims / 2.0,
- self.crop_dims / 2.0
- ])
- crop = crop.astype(int)
- input_ = input_[:, crop[0]:crop[2], crop[1]:crop[3], :]
+ #if oversample:
+ # # Generate center, corner, and mirrored crops.
+ # input_ = caffe.io.oversample(input_, self.crop_dims)
+ #else:
+ # # Take center crop.
+ # center = np.array(self.image_dims) / 2.0
+ # crop = np.tile(center, (1, 2))[0] + np.concatenate([
+ # -self.crop_dims / 2.0,
+ # self.crop_dims / 2.0
+ # ])
+ # crop = crop.astype(int)
+ # input_ = input_[:, crop[0]:crop[2], crop[1]:crop[3], :]
# Classify
caffe_in = np.zeros(np.array(input_.shape)[[0, 3, 1, 2]],
@@ -91,8 +91,8 @@ class Classifier(caffe.Net):
predictions = out[self.outputs[0]]
# For oversampling, average predictions across crops.
- if oversample:
- predictions = predictions.reshape((len(predictions) // 10, 10, -1))
- predictions = predictions.mean(1)
+ #if oversample:
+ # predictions = predictions.reshape((len(predictions) // 10, 10, -1))
+ # predictions = predictions.mean(1)
return predictions
最后,真正到运行命令的时候了:
使用命令计算图片:
python python/classify.py --model_def examples/mnist/lenet.prototxt --pretrained_model examples/mnist/lenet_iter_10000.caffemodel --center_only --images_dim 28,28 /home/user/Desktop/2.jpg FOO
更改上在py程序后,会输出以下内容 :
Saving results into FOO
[[2.63039285e-10 1.69570372e-10 1.00000000e+00 3.37297190e-10
1.04435086e-16 6.86246951e-15 1.50223258e-14 4.68932055e-12
6.54263449e-11 1.28875165e-14]]
由于输入的是2.jpg,所以第2个位置(从0开始)的概率最大,几乎是1.可以分别手写0到9图片,进行测试。
我分别在vm(i53230), j1900(真机), i737**m(真机)进行测试,cpu运行过,识别率还可以,速度感觉都比较慢在1s以上吧。