def run(self, topic=False, show=True, radar=False, ctx=gpu(0)):
self.radar = radar
self.show = show
self.topk = 1
self._init_ros()
self.resz = image.ForceResizeAug((self.size[1], self.size[0]))
if radar:
plt.ion()
fig = plt.figure()
self.ax = fig.add_subplot(111, polar=True)
self.ax.grid(True)
if not topic:
threading.Thread(target=self._get_frame).start()
print('\033[1;33;40m Use USB Camera')
else:
rospy.Subscriber(topic, Image, self._image_callback)
print('\033[1;33;40m Image Topic: %s\033[0m'%topic)
rate = rospy.Rate(30)
while not rospy.is_shutdown():
if hasattr(self, 'img'):
nd_img = nd.array(self.img)
nd_img = self.resz(nd_img).as_in_context(ctx)
nd_img = nd_img.transpose((2,0,1)).expand_dims(axis=0)/255.
out = self.predict(nd_img)
self.visualize(out)
def predict_mxnet(net, ctx, fname, label):
'''
使用mxnet对图像进行预测
:param net:训练好的模型
:param ctx:数据context
:param fname:图像路径
:param label:标签词典
:return:预测类别及概率
'''
with open(fname, 'rb') as f:
img = image.imdecode(f.read())
img = image.ForceResizeAug((image_size, image_size))(img)
data, _ = transform_test(img, -1)
data = data.expand_dims(axis=0)
out = net(data.as_in_context(ctx))
out = nd.SoftmaxActivation(out)
pred = int(nd.argmax(out, axis=1).asscalar())
prob = out[0][pred].asscalar()
return '置信度=%f, 类别 %s' % (prob, label[str(pred)])
import h5py
import os
import gc
from tqdm import tqdm
from sklearn.model_selection import train_test_split
# In[ ]:
# try run via gpu
ctx = mx.gpu()
# In[ ]:
transformers = [
# 强制resize成pretrain模型的输入大小
image.ForceResizeAug((224, 224)),
# 标准化处理
image.ColorNormalizeAug(mean=nd.array([0.485, 0.456, 0.406]),
std=nd.array([0.229, 0.224, 0.225]))
]
# Transoform image and label to our target format data
#
# 为避免过拟合，我们在这里使用`image.CreateAugmenter`来加强数据集。例如我们设`rand_mirror=True`即可随机对每张图片做镜面反转。我们也通过`mean`和`std`对彩色图像RGB三个通道分别做标准化。以下我们列举了该函数里的所有参数，这些参数都是可以调的。
# In[ ]:
train_augs_params = {
"resize": 1,
"rand_crop": False,
"rand_resize": False,
from mxnet.gluon.model_zoo import vision as models
import numpy as np
from tqdm import tqdm
import h5py
import matplotlib.pyplot as plt
#%matplotlib inline
#%config InlineBackend.figure_format = 'retina'
import warnings
warnings.filterwarnings("ignore")
ctx = mx.gpu()
preprocessing = [
image.ForceResizeAug((224, 224)),
image.ColorNormalizeAug(mean=nd.array([0.485, 0.456, 0.406]),
std=nd.array([0.229, 0.224, 0.225]))
]
def transform(data, label):
data = data.astype('float32') / 255
for pre in preprocessing:
data = pre(data)
data = nd.transpose(data, (2, 0, 1))
return data, nd.array([label]).asscalar().astype('float32')
def get_features(net, data):
def get_image_mat(data_path, pic_size):
img = image.imdecode(open(data_path, 'rb').read())
if pic_size[0]*pic_size[1] != 0:
img = image.ForceResizeAug(pic_size)(img)
return img.astype('float32').asnumpy()
from mxnet import init
from mxnet import nd
from mxnet.gluon.data import vision
from mxnet.gluon.model_zoo import vision as models
import numpy as np
from tqdm import tqdm
import h5py
import matplotlib.pyplot as plt
import warnings
warnings.filterwarnings("ignore")
ctx = mx.gpu()
preprocessing = [
image.ForceResizeAug((224, 224)),
image.ColorNormalizeAug(mean=nd.array([0.485, 0.456, 0.406]),
std=nd.array([0.229, 0.224, 0.225]))
]
def transform(data, label):
data = data.astype('float32') / 255
for pre in preprocessing:
data = pre(data)
data = nd.transpose(data, (2, 0, 1))
return data, nd.array([label]).asscalar().astype('float32')
def get_features(net, data):