def avg_perf_savefig(avg_cscore, c_score, subj_name, params=None):
""" Saves 2 figures at specified path using params key= list [0 or 1,save_path_string]
-first element of the "params" list 1 or 0 activates or deactivates plot function.
-second element is the string which specifies save path for the plot.
fig1: Class scores 1 subj vs all subs
fig2 : Average performance save fig """
if params[0] == 1:
goal_dir = os.path.join(params[1])
# plot 1 subj vs all
name1 = '1vsAll'
axis1 = ['subjects', 'subjects']
mpt.plot_confusion_matrix(c_score, classes=subj_name, normalize=False, title='mhad class score per subject', axs=axis1)
plt.savefig(goal_dir + name1)
plt.close('all')
# # Plot Average Performance in dataset
name2 = 'avg_performance_dtset'
col_label = ['average']
img_view = np.reshape(avg_cscore, (12, 1))
im, cbar= heatmap(img_view, subj_name, col_label, cmap='jet')
texts = annotate_heatmap(im, valfmt="{x:.2f} ")
plt.axes().set_aspect('auto')
plt.title('Average Performance for every Subject\n Mhad_dataset')
plt.savefig(goal_dir + name2)
plt.close('all')
def classifyKNN(frame, database, k, metric):
distances = []
for iframe in range(0, database.shape[0]):
d = metric(frame, database[iframe][0])
distances.append((d, database[iframe][1], database[iframe][2]))
distances.sort(key=lambda row: row[0])
return mpt.most_often_occurence(distances[0:k])
def cfm_savefig(subject1,subject2,params_cmf):
"""Saves Confusion matrix of 2 subjects and 11 actions at the specified path params_cmf[]
input: subject1, subject2 names strings
ouput: plt.savefig to path"""
if params_cmf[4] == 1:
goal_dir = os.path.join(params_cmf[5])
axlabel = [subject2, subject1] # [x,y]
mpt.plot_confusion_matrix(params_cmf[0], classes=params_cmf[1], normalize=False, title='confusion matrix', axs=axlabel)
# plt.plot(indexes[:,0], indexes[:,1],'ro')
plt.title('Classification Score = ' + str(round(params_cmf[2], 2)) + '%\nmisclassified=' + str(int(params_cmf[3])))
plt.rcParams.update({'font.size': 13})
plt.tight_layout()
# plt.show()
name = subject1 + '_' + subject2
my_file = name + '_cfm'
plt.savefig(goal_dir + my_file)
plt.close('all')
# DTW figures path
savefig_dtw = os.getcwd() + "/plots/MP_Similarity_Matrix/dtw_results/"
# sflag = 0 : Turn off plots , 1: save figures to path
sflag = 1
FV_new = []
for name in dataset:
dataset_dir = "/home/evangeloit/Desktop/GitBlit_Master/PythonModel3dTracker/Data/" + name + ".json"
input_dir = "/home/evangeloit/Desktop/GitBlit_Master/PythonModel3dTracker/Data/rs/Human_tracking/" \
+ name + "_results_ldm.json"
## Load data from Json ##
dataPoints, dataLim = mpt.load_data(input_dir, dataset_dir)
init_frame = 0 # dataLim['limits'][0]
last_frame = 110 # dataLim['limits'][1]
##### Create 3D points array #####
p3d = mpt.Create3dPoints(init_frame, last_frame, dataPoints, model_name)
## Gaussian Filter 5 by 1 in time dimension
p3d_gauss = mpt.GaussFilter3dPoints(p3d, sigma, t)
#### Create Feature Vector ####
feat_vec, vec, acc = mpt.MovPoseDescriptor(p3d_gauss, StartFrame)
from Moving_Pose_Descriptor import MP_tools2 as mpt
from Moving_Pose_Descriptor import confmat as cfm
from Moving_Pose_Descriptor import Threshold_Precision_Recall as TPR
# from Moving_Pose_Descriptor.heatmap import heatmap
# from Moving_Pose_Descriptor.heatmap import annotate_heatmap
# import matplotlib.pyplot as plt
# from heatmap import heatmap
# from heatmap import annotate_heatmap
# from munkres import Munkres
import json
# Controllers
#Open mhad dataset jsonfile
with open(os.path.join(os.environ['mvpd'], "dataset.json")) as f:
dataset_s1 = mpt.AlpNumSorter(list(json.load(f)))
#Paths
dtpath = '/home/evangeloit/Desktop/GitBlit_Master/PythonModel3dTracker/Data/data/'
landmarks_path = "/home/evangeloit/Desktop/GitBlit_Master/PythonModel3dTracker/Data/rs/Human_tracking/results_camera_invariant/"
# os.chdir(dtpath) # Mhad Dataset directory
model_name = 'mh_body_male_customquat'
# Actions
actions = [
"A01", "A02", "A03", "A04", "A05", "A06", "A07", "A08", "A09", "A10", "A11"
]
# Gaussian Filter Parameters
sigma = 1
# "mh_body_male_customquat"
# ]
# datasets = [
# 'mhad_s03_a04','mhad_s06_a04',
# 'mhad_s12_a04','mhad_s11_a04',
# 'mhad_s05_a04','mhad_s02_a04',
# 'mhad_s10_a04'
# ]
mod_name = ["mh_body_male_customquat"]
# datasets = dtlist.datasets_list(dtpath)
#Open mhad dataset jsonfile
with open(os.path.join(os.environ['mvpd'], "dataset.json")) as f:
datasets = mpt.AlpNumSorter(list(json.load(f)))
#take out specific actions
print(datasets)
print(len(datasets))
k = len(datasets)
model_names = mod_name * k
# print(model_names)
# print(datasets)
# print(len(model_names))
# print(len(datasets))
#model_names = ["mh_body_male_customquat"] * len(datasets)
#model_names = ["mh_body_male_custom"]
#datasets = ["mhad_ammar"]
database = np.array(database)
testframes = database.copy()
# compute confidence for every frame
k = 5
class_frames = []
for income in range(0, testframes.shape[0]):
#incoming frame Feature Vector
fv_in = testframes[income][0]
# print("incoming frame no:",income ," iSub", testframes[income][1], " iAct", testframes[income][2] )
print("frame :", income)#," iSub", testframes[income][1], " iAct", testframes[income][2] )
#filter Database
newdatabase = mpt.filterdatabase(testframes[income],database) #exclude iSubject's all iAction frames
#Random frame numbers
partData=mpt.randomData(newdatabase, percent=0.1)
#confidence of frame that belong to a class (frame_feature vector ,
confidence_tuple = FrameWiseClassify.classframe(fv_in, newdatabase, k, dataPercent=partData)
cf = (fv_in, confidence_tuple)
class_frames.append(cf)
class_frames = np.array(class_frames)
np.save('class_frames_k5.npy',class_frames)
print("--- %s seconds ---" % (time.time() - start_time))
# compute confidence for every frame
k = 4
class_frames = []
ct = 0
for income in range(0, 1100):
#incoming frame Feature Vector
fv_in = testframes[income][0]
# print("incoming frame no:",income ," iSub", testframes[income][1], " iAct", testframes[income][2] )
print(
"frame :", income
) #," iSub", testframes[income][1], " iAct", testframes[income][2] )
#filter Database
newdatabase = mpt.filterdatabase(
testframes[income], database) #exclude iSubject's all iAction frames
#Random frame numbers
partData = None #mpt.randomData(newdatabase, percent=None)
#confidence of frame that belong to a class (frame_feature vector ,
confidence_tuple = FrameWiseClassify.classframe(fv_in,
newdatabase,
k,
dataPercent=partData)
change_action = abs(testframes[income][2] - testframes[income + 1][2])
if change_action != 0:
ct = 0
def Conf2Subject(subject1,subject2,dtpath,fv_1,fv_2,params=None ):
"""Conf2Subj ::
inputs: subject1,subject2,dataset path,feature vector of s1,FV of s2,
params =[0 or 1,string] / 0: No plot saving ,1: saves all plots, string: contains the absolute path of the
destination save folder.
output: m x m score matrix between subj1,subj2 for act [0-11]
class_score : (rowmin+colmin)/(2*num of actions) * %
missclass"""
act_s1, act_s1_not = mpt.list_ext(os.path.join(dtpath, subject1), 'json')
act_s2, act_s2_not = mpt.list_ext(os.path.join(dtpath, subject2), 'json')
act_s1_not = mpt.AlpNumSorter(act_s1_not)
act_s2_not = mpt.AlpNumSorter(act_s2_not)
# print(act_s1_not)
# print(act_s2_not)
# print("new_pair:")
score = np.empty((len(act_s1_not), len(act_s2_not)), np.dtype(np.float32))
for sub1 in range(0, len(act_s1_not)):
for sub2 in range(0, len(act_s2_not)):
Y = cdist(fv_1[sub1], fv_2[sub2], 'euclidean')
p, q, C, phi = mpt.dtwC(Y, 0.1)
# Sum of diagonal steps
dsteps = 0
for r in range(0, len(q) - 1):
qdot = abs(q[r] - q[r + 1])
pdot = abs(p[r] - p[r + 1])
s = qdot + pdot
if s == 2:
dsteps = dsteps + 1
# print(dsteps)
# Scores of DTW for every subject / Objective Function
score[sub1][sub2] = (C[-1, -1] / ((Y.shape[0] + Y.shape[1])))
mpt.DistMatPlot(Y, params[1], q, p, dtwscore=score[sub1][sub2],
name=act_s1_not[sub1] + "_" + act_s2_not[sub2], flag='DTW',
save_flag=params[0])
#Class Score [min row + min col]
# Pscore = ((score/np.amax(score))*100).copy()
Pminrow = np.argmin(score, axis=1) # axis=1 row min index
Pmincol = np.argmin(score, axis=0) # axis=0 col min index
Pvec = np.arange(0, len(Pminrow))
pr = Pminrow == Pvec
pr = pr*1
pc = Pmincol == Pvec
pc = pc*1
mtot = pr + pc
missclass = (2*len(score)) - np.sum(mtot)
# # print(missclass)
class_score = (np.sum(mtot, dtype=float) / (2 * len(score))) * 100
# print(class_score)
return score, class_score , missclass
# Compare one set with all the other datasets -- save Figure path
savefig_comp = os.getcwd() + "/plots/conf_matrix/MP_comp_mat/"
# DTW figures path
savefig_dtw = os.getcwd() + "/plots/conf_matrix/dtw_res_conf/"
# sflag = 0 : Turn off plots , 1: save figures to path
sflag = 0
FV_new = []
subj_name = os.listdir(os.getcwd()) # List of Subjects in the directory
for subj in range(0, len(subj_name)): #for every subject
acts, acts_not = mpt.list_files1(
os.path.join(os.getcwd(), subj_name[subj]), 'json')
# print(acts)
# # acts = os.listdir(os.path.join(os.getcwd(), subj_name[subj]))
for act in range(0, len(acts)): # for every action of a subject
# for name in dataset_s1:
# dataset_dir = "/home/evangeloit/Desktop/GitBlit_Master/PythonModel3dTracker/Data/data/" + name + ".json"
dataset_dir = os.path.join(os.getcwd(), subj_name[subj], acts[act])
input_dir = "/home/evangeloit/Desktop/GitBlit_Master/PythonModel3dTracker/Data/rs/Human_tracking/results_camera_invariant/" \
+ acts_not[act] + "_ldm.json"
## Load data from Json ##
dataPoints, dataLim = mpt.load_data(input_dir, dataset_dir)
init_frame = dataLim['limits'][0]
last_frame = dataLim['limits'][1]