def myBuildDBNtest(self):
#提取所有帧中，原始的骨架点，得到一个矩阵，Skeleton_matrix ，同时返回骨架是否归0化
#Skeleton_matrix, valid_skel = Extract_feature_UNnormalized(smp, used_joints, 1, smp.getNumFrames())
time_tic = time.time()
import cPickle
Skeleton_matrix=cPickle.load(open("testSkeleton_matrix","rb"))
#print Skeleton_matrix
Feature = Extract_feature_Realtime(Skeleton_matrix, self.njoints)
Feature_normalized = normalize(Feature, self.Mean1, self.Std1)
'''
##########################
### model 1 第一种网络构架模式 #
##########################
dbn = GRBM_DBN(numpy_rng=numpy_rng, n_ins=528,
hidden_layers_sizes=[1000, 1000, 500],
n_outs=201)
dbn.load('dbn_2014-05-23-20-07-28.npy')#预先训练好的构架
#这里就是theano的奇葩函数构架
validate_model = theano.function(inputs=[],
outputs=dbn.logLayer.p_y_given_x,#输出是逻辑回归层的输出
givens={ dbn.x: shared_x}) '''
observ_likelihood_1 = self.validate_model(Feature_normalized)#调用函数得到结果
##########################
# viterbi path decoding
#####################
observ_likelihood_1=observ_likelihood_1[0:50,:]
#这里自己改了，只有第一个网络结构的，
log_observ_likelihood = log(observ_likelihood_1.T)
#这里是一个矩阵， 行是样本，列是概率 [1884个样本, 201列] 用T转置了
print "处理时间 %d sec" % int(time.time() - time_tic)
time_tic = time.time()
#下面就是vibiter算法了
print("\t Viterbi path decoding " )
# do it in log space avoid numeric underflow
[path, predecessor_state_index, global_score] =viterbi_path_log(
self.Prior, self.Transition_matrix, log_observ_likelihood )
label=viterbi_endframe(path,5,30)
# Some gestures are not within the vocabulary
#[pred_label, begin_frame, end_frame, Individual_score, frame_length] = viterbi_colab_clean(
# path, global_score, threshold=-100, mini_frame=19)
print "标记是:"
print label
print "viterbi处理时间 %d sec" % int(time.time() - time_tic)
def myBuildDBN(self,Skeleton_matrix):
#提取所有帧中，原始的骨架点，得到一个矩阵，Skeleton_matrix ，同时返回骨架是否归0化
#Skeleton_matrix, valid_skel = Extract_feature_UNnormalized(smp, used_joints, 1, smp.getNumFrames())
time_tic = time.time()
Feature = Extract_feature_Realtime(Skeleton_matrix, self.njoints)
Feature_normalized = normalize(Feature, self.Mean1, self.Std1)
observ_likelihood_1 = self.validate_model(Feature_normalized)#调用函数得到结果
##########################
# viterbi path decoding
#####################
#observ_likelihood_1=observ_likelihood_1[0:50,:]
#这里自己改了，只有第一个网络结构的，
log_observ_likelihood = log(observ_likelihood_1.T)
#这里是一个矩阵， 行是样本，列是概率 [1884个样本, 201列] 用T转置了
print "处理时间 %d sec" % int(time.time() - time_tic)
return log_observ_likelihood
if file_count<800:
print("\t Processing file " + file)
# Create the object to access the sample
smp=GestureSample(os.path.join(data,file))
# ###############################################
# USE Ground Truth information to learn the model
# ###############################################
# Get the list of actions for this frame
gesturesList=smp.getGestures()
frame_num = smp.getNumFrames()
Feature_Array = np.zeros(shape = (frame_num , (njoints*(njoints-1)/2 + njoints**2)*3),dtype=np.float32)
# Target = np.zeros( shape=(frame_num, target_category), dtype=np.uint8)
#feature generate
Skeleton_matrix, valid_skel = Extract_feature_UNnormalized(smp,used_joints, 1, frame_num)
Feature_Array = Extract_feature_Realtime(Skeleton_matrix, njoints)
Feature_all.append(Feature_Array)
#target generate
labels = np.zeros(frame_num, np.uint8)
for row in gesturesList:
labels[int(row[1])-1:int(row[2])-1] = int(row[0])
Target_all.append(labels)
del smp
# save the skeleton file:
import cPickle as pickle
f = open('Feature_train_realtime.pkl','wb')
def generate_eigenjoint(feature_name='sk_eigenjoint_nor_528',
labels_name='labels_raw'):
# Data folder (Training data)
print("Extracting the training files")
data = os.path.join("E:\\program\\Chalearn\\rawdata\\train\\")
target_dir = 'E:\\program\\Chalearn\\Chalearn_LSTM\\target\\'
# Get the list of training samples
samples = os.listdir(data)
output_dir = 'E:\\program\\Chalearn\\Chalearn_LSTM\\feature\\' + feature_name
if not os.path.exists(output_dir):
os.makedirs(output_dir)
used_joints = [
'ElbowLeft', 'WristLeft', 'ShoulderLeft', 'HandLeft', 'ElbowRight',
'WristRight', 'ShoulderRight', 'HandRight', 'Head', 'Spine',
'HipCenter'
]
njoints = len(used_joints)
f = open('SK_normalization.pkl', 'r')
normal_params = pickle.load(f)
f.close()
Mean = normal_params['Mean1']
Std = normal_params['Std1']
count = 0
# target_category = 21
Target_all = []
#Feature_all = numpy.zeros(shape=(400000, (njoints*(njoints-1)/2 + njoints**2)*3),dtype=numpy.float32)
for file_count, file in enumerate(samples):
if int(file[-8:-4]) != 417 and int(file[-8:-4]) != 675:
print("\t Processing file " + file)
# Create the object to access the sample
smp = GestureSample(os.path.join(data, file))
# ###############################################
# USE Ground Truth information to learn the model
# ###############################################
# Get the list of actions for this frame
gesturesList = smp.getGestures()
frame_num = smp.getNumFrames()
Feature_Array = np.zeros(
shape=(frame_num,
(njoints * (njoints - 1) / 2 + njoints**2) * 3),
dtype=np.float32)
# Target = np.zeros( shape=(frame_num, target_category), dtype=np.uint8)
#feature generate
Skeleton_matrix, valid_skel = Extract_feature_UNnormalized(
smp, used_joints, 1, frame_num)
Feature_Array = Extract_feature_Realtime(Skeleton_matrix, njoints)
Feature_Array = normalize(Feature_Array, Mean, Std)
add_ = Feature_Array[-1].reshape((1, Feature_Array.shape[1]))
Feature_Array = np.concatenate((Feature_Array, add_), axis=0)
#save sample sk features
output_name = '%04d.npy' % count
count += 1
np.save(os.path.join(output_dir, output_name), Feature_Array)
#target generate
labels = np.zeros(frame_num, np.uint8)
for row in gesturesList:
labels[int(row[1]) - 1:int(row[2]) - 1] = int(row[0])
Target_all.append(labels)
del smp
np.save(target_dir + '%s.npy' % labels_name, Target_all)
# Iterate for each action in this sample
for gesture in gesturesList:
#对于每个手势数据
# Get the gesture ID, and start and end frames for the gesture
gestureID,startFrame,endFrame=gesture#每个手势包含了 手势的ID ， 开始结束帧的序号。
Skeleton_matrix, valid_skel = Extract_feature_UNnormalized(smp, used_joints, startFrame, endFrame)
#骨架的节点坐标矩阵。 是否骨架归一化了（中心位置）
# to see we actually detect a skeleton:
if not valid_skel:
print "No detected Skeleton: ", gestureID
else:
### extract the features according to the CVPR2014 paper
Feature = Extract_feature_Realtime(Skeleton_matrix, njoints)##提取特征，这里只有当前帧的坐标差， 和前后帧坐标差 FC CC
Target = numpy.zeros( shape=(Feature.shape[0], STATE_NO*20+1))
#hmm中20个动作 每个都有10个隐含状态。
fr_no = Feature.shape[0]#有多少个特征值
for i in range(STATE_NO): #HMM states force alignment 这里是强制将动作帧分成10份，每个作为一个状态
begin_fr = numpy.round(fr_no* i /STATE_NO) + 1
end_fr = numpy.round( fr_no*(i+1) /STATE_NO)
#print "begin: %d, end: %d"%(begin_fr-1, end_fr)
seg_length=end_fr-begin_fr + 1
targets = numpy.zeros( shape =(STATE_NO*20+1,1))#201 行，1列 的结果矩阵
targets[ i + STATE_NO*(gestureID-1)] = 1#对应这个动作*20+隐含状态序号==》标签的全局序号赋值为1
begin_frame = count
end_frame = count+seg_length
Feature_all[begin_frame:end_frame,:] = Feature[begin_fr-1:end_fr,:]#将特征拼接起来