def ofclipcustom(clippath, lkparams):
#load gt for init
gt = h.readmaarray(clippath + "/groundtruth/gt")
gtcor = gt[:,2:6]
gtcen = h.corcen(gt[:,2:6])
#new masked array to remember of predictions
memof = ma.array(np.zeros((np.shape(gt)[0], 4, 2)), mask = True, dtype = np.float32)
memof[0,0,:] = gtcor[0,:2]
memof[0,1,:] = [gtcor[0,2]-5, gtcor[0,1]]
memof[0,2,:] = [gtcor[0,0], gtcen[0,1]]
memof[0,3,:] = gtcen[0,:2]
memstatus = np.zeros((np.shape(gt)[0],4))
for i in range(1,gt.shape[0]):
oldframe = cv2.imread(clippath + "/data/"+ h.extract(clippath) + "#" + str(i-1) + ".png",1)
nowframe = cv2.imread(clippath + "/data/"+ h.extract(clippath) + "#" + str(i) + ".png",1)
for j in range(memof.shape[1]):
newpoints, status = ofupdate(oldframe, nowframe, memof[i-1,j,:], lkparams)
#newpoints, st, err = cv2.calcOpticalFlowPyrLK(oldframe,nowframe, oldpoints[0,:,:,:],None, **lkparams)
memof[i,j,:] = newpoints
memstatus[i,j] = status
if status == 0:
memof.mask[i,j,:] = True
return memof, memstatus
def ofclip(clippath, lkparams):
#load gt for init
gt = h.readmaarray(clippath + "/groundtruth/gt")
gt = h.corcen(gt[:,2:6])
#new masked array to remember of predictions
memof = ma.array(np.zeros((np.shape(gt)[0], 2)), mask = True, dtype = np.float32)
memof[0,:] = gt[0,:2]
for i in range(1,gt.shape[0]):
oldframe = cv2.imread(clippath + "/data/"+ h.extract(clippath) + "#" + str(i-1) + ".png",1)
nowframe = cv2.imread(clippath + "/data/"+ h.extract(clippath) + "#" + str(i) + ".png",1)
mem_of = np.zeros((3,2,1,2), dtype = np.float32)
mem_of[0,:,:,:] = np.array([[(3,4)],[(1,2)]])
oldpoints = np.array([[[memof[i-1,:]]]], dtype = np.float32)
newpoints = ofupdate(oldframe, nowframe, memof[i-1,:], lkparams)
#newpoints, st, err = cv2.calcOpticalFlowPyrLK(oldframe,nowframe, oldpoints[0,:,:,:],None, **lkparams)
memof[i,:] = newpoints
return memof
def handle(self, input):
result = self.pattern.match(input)
self.script_name = result.group(1)
self.regex = result.group(2)
file_contents = helper.load(self.script_name)
self.result =helper.extract(file_contents, self.regex)
def visof(points,clippath, savepath):
video = cv2.VideoWriter(savepath + ".avi",cv2.VideoWriter_fourcc(*"MJPG"),2,(1920,1080))
for i in range(points.shape[0]):
img = cv2.imread(clippath + "/data/"+ h.extract(clippath) + "#" + str(i) + ".png",1)
if points.mask[i, 0] == False:
img = cv2.circle(img, tuple(map(int,(points[i]))), 2, (255,0,255), 10)
cv2.putText(img, "Ground Truth", (10,700), cv2.FONT_HERSHEY_SIMPLEX, 3, (255,0,255),4)
video.write(img)
video.release()
def visofcustom(points,clippath, savepath):
video = cv2.VideoWriter(savepath + ".avi",cv2.VideoWriter_fourcc(*"MJPG"),2,(1920,1080))
for i in range(points.shape[0]):
img = cv2.imread(clippath + "/data/"+ h.extract(clippath) + "#" + str(i) + ".png",1)
for j in range(points.shape[1]):
if points.mask[i,j, 0] == False:
img = cv2.circle(img, tuple(map(int,(points[i,j]))), 2, (255,0,255), 3)
cv2.putText(img, "optical Flow", (10,700), cv2.FONT_HERSHEY_SIMPLEX, 3, (255,0,255),4)
cv2.putText(img, "Frame: " + str(i), (10,100), cv2.FONT_HERSHEY_SIMPLEX, 3, (0,255,0),4)
video.write(img)
video.release()
def ofcustompoint(clippath, point,gt, lkparams):
#new masked array to remember of predictions
memof = ma.array(np.zeros((np.shape(gt)[0], 2)), mask = True, dtype = np.float32)
memof[0,:] = point
for i in range(1,gt.shape[0]):
oldframe = cv2.imread(clippath + "/data/"+ h.extract(clippath) + "#" + str(i-1) + ".png",1)
nowframe = cv2.imread(clippath + "/data/"+ h.extract(clippath) + "#" + str(i) + ".png",1)
oldpoints = np.array([[[memof[i-1,:]]]], dtype = np.float32)
newpoints,status = ofupdate(oldframe, nowframe, memof[i-1,:], lkparams)
#newpoints, st, err = cv2.calcOpticalFlowPyrLK(oldframe,nowframe, oldpoints[0,:,:,:],None, **lkparams)
if status == 0:
memof[i:,:] = memof[i-1,:]
return memof
memof[i,:] = newpoints
return memof
def phrases(filename):
#fl = open('sample-output.txt','r')
fl = open(filename,'r')
megaList = []
for line in fl:
megaList.append(line)
fl.close()
lastLine = len(megaList)-1
s2p = OrderedDict() # sentences to phrases mapping
p2m = OrderedDict() # phrases to metamaps mapping
# We are creating a mapping of indexes of processing to phrases -> senetences to phrases
# take care if phrase is like 'and' or its last phrase
for i in range(len(megaList)):
if megaList[i][0:10] == 'Processing':
s2p[i] = []
currentSentencePointer = i
if megaList[i][0:6] == 'Phrase':
currentPhasePointer = i
s2p[currentSentencePointer].append(currentPhasePointer)
for k,v in s2p.items():
if len(v) > 0:
for i in range(len(v)):
if i == len(v)-1: #last phrase
p2m[v[i]] = []
j = v[i]
while j <= lastLine:
if megaList[j][0:10] == 'Processing':
break
if megaList[j][0:12] == 'Meta Mapping':
p2m[v[i]].append(j)
j = j+ 1
else: #not last phrase
p2m[v[i]] = []
for j in range(v[i]+1,v[i+1]):
if megaList[j][0:12] == 'Meta Mapping':
p2m[v[i]].append(j)
findings_with_contexts = helper.extract(s2p,p2m,megaList)
return findings_with_contexts
def drawsinglebbox(clippath, framenumber, savepath):
i = framenumber
img = cv2.imread(clippath + "/data/"+ h.extract(clippath) + "#" + str(i) + ".png",1)
bbox = cv2.selectROI(img)
bbox = list(bbox)
bbox[2] = bbox[0] + bbox[2]
bbox[3] = bbox[1] + bbox[3]
img = cv2.rectangle(img,(int(bbox[0]),int(bbox[1])),(int(bbox[2]),int(bbox[3])), (255,255,0),3) #yolo
cv2.imwrite(savepath + ".png", img)
def customimagesave(clippath, framenumber, savepath):
bbox = h.readmaarray(clippath + "/detections/raw")
i = framenumber
img = cv2.imread(clippath + "/data/"+ h.extract(clippath) + "#" + str(i) + ".png",1)
for j in range(bbox.shape[1]//6):
if bbox.mask[i, j*6] == False:
img = cv2.rectangle(img,(int(bbox[i,j*6+2]),int(bbox[i,j*6+3])),(int(bbox[i,j*6 + 4]),int(bbox[i,j*6 + 5])), (255,255,0),3) #yolo
cv2.imwrite(savepath + ".png", img)
def drawgoodpoints(clippath, framenumber, savepath):
feature_params = dict( maxCorners = 100,
qualityLevel = 0.3,
minDistance = 7,
blockSize = 7 )
i = framenumber
img = cv2.imread(clippath + "/data/"+ h.extract(clippath) + "#" + str(i) + ".png",1)
imggray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
points = cv2.goodFeaturesToTrack(imggray, mask = None, **feature_params)
for i in range(points.shape[0]):
imgpoints = cv2.circle(img, tuple(map(int,(points[i,0,:]))), 2, (255,0,255), 2)
#img = cv2.circle(img,(int(bbox[0]),int(bbox[1])),(int(bbox[2]),int(bbox[3])), (255,255,0),3) #yolo
cv2.imwrite(savepath + ".png", imgpoints)
def kalclip(bbox, gt, model, clippath, stats, ofpoint=0, R=500, Q=500, P=0):
#define start(no need to initiate before detections)
start = h.getstartindex(bbox)
clipname = h.extract(clippath)
#for 30fps --> 1/fps
#irrelevant in my case --> set 1 (i dont rly make a reference to reality here)
deltat = 1
#define what model has what representation for retransformation after:
correp = []
cenrep = ["cenof", "simplecen"]
asprep = ["aspof", "aspofwidth", "aspofwidthman"]
std = 70
#define dimensions and model(depends on representation)
if model == "simplecen":
#transform representation
bbox = h.corcen(bbox)
gt = h.corcen(gt)
#set parameters
n_x, n_z = 4, 4
Kal = KalmanFilter(dim_x=n_x, dim_z=n_z)
Kal.F = np.eye(n_x)
Kal.H = np.zeros((n_z, n_x))
Kal.H[0][0] = Kal.H[1][1] = Kal.H[2][2] = Kal.H[3][3] = 1
Kal.Q = np.eye(n_x) * std**2
for i in range(len(stats)):
Kal.R[i, i] = stats[0, 1, i]**2
#init
init = np.zeros(n_x)
init[:] = gt[0, :]
memp = np.zeros((bbox.shape[0], n_x))
memk = np.zeros((bbox.shape[0], n_x))
elif model == "cenof":
#transform representation
bbox = h.corcen(bbox)
gt = h.corcen(gt)
#set parameters
n_x, n_z, n_u = 4, 4, 2
Kal = KalmanFilter(dim_x=n_x, dim_z=n_z, dim_u=n_u)
Kal.R = np.eye(n_z) * R
for i in range(len(stats)):
Kal.R[i, i] = stats[1, 1, i]**2
Kal.Q = np.eye(n_x) * std**2
Kal.Q[0, 0], Kal.Q[1, 1] = stats[3, 1, 0], stats[3, 1, 1]
Kal.P *= P
Kal.F = np.eye(n_x)
Kal.H = np.zeros((n_z, n_x))
Kal.H[0, 0] = Kal.H[1, 1] = Kal.H[2, 2] = Kal.H[3, 3] = 1
Kal.B = np.eye(n_x, n_u)
#init
init = np.zeros(n_x)
init[:4] = gt[0, :]
memp = np.zeros((bbox.shape[0], n_x))
memk = np.zeros((bbox.shape[0], n_x))
elif model == "aspof":
#transform representation
bbox = h.corasp(bbox)
gt = h.corasp(gt)
#set parameters
n_x, n_z, n_u = 4, 4, 2
Kal = KalmanFilter(dim_x=n_x, dim_z=n_z, dim_u=n_u)
Kal.R = np.eye(n_z) * R
for i in range(len(stats)):
Kal.R[i, i] = stats[2, 1, i]**2
Kal.Q = np.eye(n_x) * std**2
Kal.Q[0, 0], Kal.Q[1, 1], Kal.Q[3,
3] = stats[3, 1,
0]**2, stats[3, 1,
1]**2, 0.0001
Kal.P *= P
Kal.F = np.eye(n_x)
Kal.B = np.eye(n_x, n_u)
Kal.H = np.zeros((n_z, n_x))
Kal.H[0, 0] = Kal.H[1, 1] = Kal.H[2, 2] = Kal.H[3, 3] = 1
#init
init = np.zeros(n_x)
init[:4] = gt[0, :]
memp = np.zeros((bbox.shape[0], n_x))
memk = np.zeros((bbox.shape[0], n_x))
elif model == "aspofwidth":
#transform representation
bbox = h.corasp(bbox)
gt = h.corasp(gt)
#set parameters
n_x, n_z, n_u = 4, 4, 3
Kal = KalmanFilter(dim_x=n_x, dim_z=n_z, dim_u=n_u)
Kal.R = np.eye(n_z) * R
for i in range(len(stats)):
Kal.R[i, i] = stats[2, 1, i]**2
Kal.Q = np.eye(n_x) * std**2
#TODO put in stats for std width from of
Kal.Q[0, 0], Kal.Q[1, 1], Kal.Q[2, 2], Kal.Q[3, 3] = stats[
3, 1, 0]**2, stats[3, 1, 1]**2, stats[3, 1, 2]**2, 0.0001
Kal.P *= P
Kal.F = np.eye(n_x)
Kal.B = np.eye(n_x, n_u)
Kal.H = np.zeros((n_z, n_x))
Kal.H[0, 0] = Kal.H[1, 1] = Kal.H[2, 2] = Kal.H[3, 3] = 1
#init
init = np.zeros(n_x)
init[:4] = gt[0, :]
memp = np.zeros((bbox.shape[0], n_x))
memk = np.zeros((bbox.shape[0], n_x))
elif model == "aspofwidthman":
#transform representation
bbox = h.corasp(bbox)
gt = h.corasp(gt)
#set parameters
n_x, n_z, n_u = 4, 4, 3
Kal = KalmanFilter(dim_x=n_x, dim_z=n_z, dim_u=n_u)
Kal.R = np.eye(n_z) * R
for i in range(len(stats)):
Kal.R[i, i] = stats[2, 1, i]**2
Kal.Q = np.eye(n_x) * std**2
Kal.Q[0, 0], Kal.Q[1, 1], Kal.Q[2, 2], Kal.Q[3, 3] = stats[
3, 1, 0], stats[3, 1, 1], 1**2, 0.0001
Kal.P *= P
Kal.F = np.eye(n_x)
Kal.B = np.eye(n_x, n_u)
Kal.H = np.zeros((n_z, n_x))
Kal.H[0, 0] = Kal.H[1, 1] = Kal.H[2, 2] = Kal.H[3, 3] = 1
#init
init = np.zeros(n_x)
init[:4] = gt[0, :]
memp = np.zeros((bbox.shape[0], n_x))
memk = np.zeros((bbox.shape[0], n_x))
#parameters to define(try these on all representations)
Kal.x = init
#create new array for results of the algorithm
mem = ma.array(np.zeros_like(bbox), mask=True)
mem[0, :] = init
for j in range(memp.shape[1]):
memp[0, j] = Kal.P[j, j]
for j in range(memk.shape[1]):
memk[0, j] = Kal.K[j, j]
for i in range(1, bbox.shape[0], 1):
z = bbox[i, :]
if bbox.mask[i, 0] == True:
z = None
if model == "cenof":
vx = ofpoint[i, 0] - ofpoint[i - 1, 0]
vy = ofpoint[i, 1] - ofpoint[i - 1, 1]
u = np.array([vx, vy])
Kal.predict(u)
elif model == "aspof":
vx = ofpoint[i, 0] - ofpoint[i - 1, 0]
vy = ofpoint[i, 1] - ofpoint[i - 1, 1]
u = np.array([vx, vy])
Kal.predict(u)
elif model == "aspofwidth":
vx = ofpoint[i, 0] - ofpoint[i - 1, 0]
vy = ofpoint[i, 1] - ofpoint[i - 1, 1]
vwidth = ofpoint[i, 2] - ofpoint[i - 1, 2]
u = np.array([vx, vy, vwidth])
Kal.predict(u)
elif model == "aspofwidthman":
vx = ofpoint[i, 0] - ofpoint[i - 1, 0]
vy = ofpoint[i, 1] - ofpoint[i - 1, 1]
vwidth = ofpoint[i, 2] - ofpoint[i - 1, 2]
u = np.array([vx, vy, vwidth])
Kal.predict(u)
else:
Kal.predict()
Kal.update(z)
x = Kal.x
y = x[:4]
#write into resulting array
mem[i, :] = y[:4]
for j in range(memp.shape[1]):
memp[i, j] = Kal.P[j, j]
for j in range(memk.shape[1]):
memk[i, j] = Kal.K[j, j]
#just return result --> do back-transformation into other representations outside of this function(as well as creating visuals)
if model in correp:
mem = mem
elif model in cenrep:
mem = h.cencor(mem)
elif model in asprep:
mem = h.aspcor(mem)
return mem, memp, memk
def main(skip):
#read info about clips and define sourcevideo name from its path(-.mp4)
mainvidpath, info = h.readtimes(textpath)
mainvid = mainvidpath.split(".")[-2]
mainvid = mainvid.split("/")[-1]
datapath = "./data/"
macropath = datapath + mainvid + "/macroanalysis/"
if skip < 1:
#delete all existing analysis and alorithm folders(so that if i changed name of some files the old ones get deleted)
h.delfolders(mainvid)
"""
####################################################################
load raw data
####################################################################
"""
#loop over clips --> create folders for clips and cut them up (if u change clip length but not name u need to delete the folder for that clip, otherwise it will just be skipped and not updated)
for i in range(np.shape(info)[0]):
#define clippath to give different methods
clippath = datapath + mainvid + "/" + info[i][2]
clipname = h.extract(clippath)
if not os.path.exists(clippath):
#create folders and cut main vid
print("creating folders and cutting: " + info[i][2])
h.clipfolder(mainvid, info[i][2])
if not os.path.exists(clippath + "/data/sourceinfo.txt"):
h.cutmainvid(mainvid, info[i][0], info[i][1], info[i][2],
mainvidpath)
if not os.path.exists(clippath + "/detections/raw.txt"):
#use yolo to get bbox of current clipfilename
print("yolo running on: " + info[i][2])
bbox = h.clipyolodet(clippath)
print("writing detections to file...")
h.writemaarray(bbox, clippath + "/detections/raw",
"rawdetectiontestheader")
#create folders
h.makedirsx(clippath + "/analysis")
h.makedirsx(clippath + "/algorithms")
h.makedirsx(clippath + "/algorithms/kalman")
h.makedirsx(clippath + "/algorithms/fm")
h.makedirsx(clippath + "/algorithms/of")
#create folder for pre analysis
h.makedirsx(clippath + "/analysis/pre")
h.makedirsx(clippath + "/analysis/post")
#create folder to store histograms over all clips errors in
h.makedirsx(macropath)
h.makedirsx(macropath + "/pre")
h.makedirsx(macropath + "/post")
#start new loop to self label all after each other, not for every single clip(would have big pauses between clips where yolo runs)
for i in range(np.shape(info)[0]):
clippath = datapath + mainvid + "/" + info[i][2]
clipname = h.extract(clippath)
#selflabel until object is going (even partly) out of frame to focus on movement of object
while (not os.path.exists(clippath + "/groundtruth/gt.txt")):
gt = h.selflabel(clippath, 0)
h.writemaarray(gt, clippath + "/groundtruth/gt",
"this is the self labeled ground truth")
print("loaded raw data + saved detections and groundtruth")
"""
####################################################################
clean data
####################################################################
"""
#clean data
for i in range(np.shape(info)[0]):
clippath = (datapath + mainvid + "/" + info[i][2])
clipname = h.extract(clippath)
if not os.path.exists(clippath + "/detections/clean.txt"):
#read gt and raw detections from text file
gt = h.readmaarray(
clippath +
"/groundtruth/gt") #need class and conf for cleanup
bboxraw = h.readmaarray(
clippath +
"/detections/raw") #need class and conf for cleanup
#clean data
bboxclean = h.cleanup(bboxraw, 0, gt)
#write clean data back
h.writemaarray(bboxclean, clippath + "/detections/clean",
"cleaned up detections")
print("cleaned data from: " + clipname)
"""
############################################################################
run opical flow
############################################################################
"""
#define windowsizes outside to better retrieve file later (if u dont want to run this section all the time)
windows = [3, 9, 27]
if skip < 2:
#define of parameters
lkparams = dict(winSize=(3, 3),
maxLevel=2,
criteria=(cv2.TERM_CRITERIA_EPS
| cv2.TERM_CRITERIA_COUNT, 10, 0.03))
#run optical flow on clips with center as init
for i in range(np.shape(info)[0]):
clippath = datapath + mainvid + "/" + info[i][2]
clipname = h.extract(clippath)
#find good points to track and save them
framenbr = 0
h.drawgoodpoints(clippath, framenbr,
clippath + "/algorithms/of/goodpoints")
#run optical flow
gt = h.readmaarray(clippath + "/groundtruth/gt")[:, 2:6]
gtcen = h.corcen(gt)
#define points to track
pointinfo = np.array([["ofcenter", gtcen[0, 0], gtcen[0, 1]],
["oftopleft", gt[0, 0], gt[0, 1]],
["oftopright", gt[0, 2], gt[0, 1]]])
for k in range(len(windows)):
lkparams["winSize"] = (windows[k], windows[k])
for j in range(pointinfo.shape[0]):
memof = of.ofcustompoint(
clippath, [pointinfo[j, 1], pointinfo[j, 2]], gt,
lkparams)
h.writemaarray(
memof, clippath + "/algorithms/of/" + clipname +
pointinfo[j, 0] + "win" + str(windows[k]),
"of initialized with " + pointinfo[j, 0] + "of gt")
of.visof(
memof, clippath, clippath + "/algorithms/of/" +
clipname + pointinfo[j, 0] + "win" + str(windows[k]))
#save of center and width in one txt file to better retrieve it later
#load 3 points
ofcenter = h.readmaarray(clippath + "/algorithms/of/" +
clipname + pointinfo[0, 0] + "win" +
str(windows[k]))
oftopleft = h.readmaarray(clippath + "/algorithms/of/" +
clipname + pointinfo[1, 0] + "win" +
str(windows[k]))
oftopright = h.readmaarray(clippath + "/algorithms/of/" +
clipname + pointinfo[2, 0] + "win" +
str(windows[k]))
#put width into from 2 points into ofpoints array and later give whole array with(centerx, centery, width) to kalman
maarray = ma.array(np.zeros((ofcenter.shape[0], 3)),
mask=True,
dtype=np.float64)
maarray[:, :2] = ofcenter
for j in range(ofcenter.shape[0]):
maarray[j, 2] = oftopright[j, 0] - oftopleft[j, 0]
ofpoints = maarray
#save in txt as (centerx,centery, width)
h.writemaarray(
ofpoints, clippath + "/algorithms/of/" + clipname +
"ofcenwidth" + "win" + str(windows[k]),
"of initialized with centerx, centery, width(top right and top left corner) of bbox(gt)"
)
print("ran optical flow on: " + clipname + " winSize: " +
str(windows[k]) + "*" + str(windows[k]))
print("done with optical flow")
"""
############################################################################
pre-analysis
############################################################################
"""
#not being used at the moment
binsize = 3
if skip < 3:
#analysis
for i in range(np.shape(info)[0]):
clippath = datapath + mainvid + "/" + info[i][2]
clipname = h.extract(clippath)
#over time
#read bboxes
bboxclean = h.readmaarray(clippath + "/detections/clean")[:, 2:6]
gt = h.readmaarray(clippath + "/groundtruth/gt")[:, 2:6]
#corner representation
#plot and save yolo bboxes
ylabel = ["xmin [px]", "ymin [px]", "xmax [px]", "ymax [px]"]
#plot and save gt bboxes
h.timeplot2(gt, bboxclean, ylabel, ["Frame [k]"] * 4,
clippath + "/analysis/pre/" + clipname + "otyologt")
#center representation
#plot and save yolo bboxes
ylabel = [
"Mittelpunktx [px]", "Mittelpunkty [px]", "Breite [px]",
"Höhe [px]"
]
#plot and save gt bboxes
h.timeplot2(h.corcen(gt), h.corcen(bboxclean), ylabel,
["Frame [k]"] * 4,
clippath + "/analysis/pre/" + clipname + "otyologt")
#aspect ratio representation
#plot and save yolo bboxes
ylabel = [
"Mittelpunktx [px]", "Mittelpunkty [px]", "Breite [px]",
"Seitenverhältnis"
]
#plot and save gt bboxes
h.timeplot2(h.corasp(gt), h.corasp(bboxclean), ylabel,
["Frame [k]"] * 4,
clippath + "/analysis/pre/" + clipname + "otyologt")
for k in range(len(windows)):
ofcenwidth = h.readmaarray(clippath + "/algorithms/of/" +
clipname + "ofcenwidth" + "win" +
str(windows[k]))
ofcenright = h.readmaarray(clippath + "/algorithms/of/" +
clipname + "oftopright" + "win" +
str(windows[k]))
ofcenleft = h.readmaarray(clippath + "/algorithms/of/" +
clipname + "oftopleft" + "win" +
str(windows[k]))
ofleftright = h.connectleftright(ofcenleft, ofcenright)
ofpointsgt = h.createofgt(gt)
ylabel = [
"Mittelpunktx [px]", "Mittelpunkty [px]", "Breite [px]",
"Seitenverhältnis"
]
#compare ofcenter and width with gt center and width
h.timeplot2(
h.corcen(gt)[:, :3], ofcenwidth, ylabel, ["Frame [k]"] * 4,
clippath + "/analysis/pre/" + clipname + "otofgt" + "win" +
str(windows[k]), ["green", "black"],
["Ground Truth", "Optischer Fluss"])
#plot of points over time compared with gt
ylabel = [
"linksobenx [px]", "linksobeny [px]", "rechtsobenx [px]",
"rechtsobeny [px]"
]
h.timeplot2(
ofpointsgt, ofleftright, ylabel, ["Frame [k]"] * 4,
clippath + "/analysis/pre/" + clipname + "otofgt" + "win" +
str(windows[k]), ["green", "black"],
["Ground Truth", "Optischer Fluss"])
#plot errors and rmse over time for every clip and every winsize
ylabel = [
"Mittelpunktx Fehler [px]", "Mittelpunkty Fehler [px]",
"Breite Fehler [px]", "Höhe Fehler [px]"
]
h.errorplot(
h.corcen(gt)[:, :3], ofcenwidth, ylabel, ["Frame [k]"] * 4,
clippath + "/analysis/pre/" + clipname + "oferror" +
"win" + str(windows[k]))
h.rmseplot(
h.corcen(gt)[:, :3], ofcenwidth, ylabel, ["Frame [k]"] * 4,
clippath + "/analysis/pre/" + clipname + "ofrmse" + "win" +
str(windows[k]))
np.savetxt(clippath + "/analysis/pre/" + clipname +
"ofrmsecw" + "win" + str(windows[k]) + ".txt",
(h.calcrmse(h.corcen(gt)[:, :3], ofcenwidth)),
fmt="%1.2f",
header="rmse for center(x,y) and width")
ylabel = [
"linksobenx RMSE [px]", "linksobeny RMSE [px]",
"rechtsobenx RMSE [px]", "rechtsobeny RMSE [px]"
]
h.errorplot(
ofpointsgt, ofleftright, ylabel, ["Frame [k]"] * 4,
clippath + "/analysis/pre/" + clipname + "oferror" +
"win" + str(windows[k]))
h.rmseplot(
ofpointsgt, ofleftright, ylabel, ["Frame [k]"] * 4,
clippath + "/analysis/pre/" + clipname + "ofrmse" + "win" +
str(windows[k]))
#save rmse as txt
np.savetxt(clippath + "/analysis/pre/" + clipname +
"ofrmselr" + "win" + str(windows[k]) + ".txt",
(h.calcrmse(ofpointsgt, ofleftright)),
fmt="%1.2f",
header="rmse for topleft(x,y) and topright(x,y)")
print("plotted over time:" + clipname)
#plot histograms for the 3 of points and the width --> for every clip and overall (+for every window size)
for k in range(len(windows)):
totalerrorofcen = []
totalerroroflr = []
for i in range(np.shape(info)[0]):
clippath = (datapath + mainvid + "/" + info[i][2])
clipname = h.extract(clippath)
bboxclean = h.readmaarray(clippath + "/detections/clean")[:,
2:6]
gt = h.readmaarray(clippath + "/groundtruth/gt")[:, 2:6]
#paths need to be correct for point nicknames of optical flow points(pointinfo above)
ofcenwidth = h.readmaarray(clippath + "/algorithms/of/" +
clipname + "ofcenwidth" + "win" +
str(windows[k]))
gtof = h.corcen(gt)[:, :3]
ofcenright = h.readmaarray(clippath + "/algorithms/of/" +
clipname + "oftopright" + "win" +
str(windows[k]))
ofcenleft = h.readmaarray(clippath + "/algorithms/of/" +
clipname + "oftopleft" + "win" +
str(windows[k]))
ofleftright = h.connectleftright(ofcenleft, ofcenright)
ofpointsgt = h.createofgt(gt)
errorof = h.error(ofcenwidth, gtof)
totalerrorofcen = h.apperror(errorof, totalerrorofcen)
erroroflr = h.error(ofleftright, ofpointsgt)
totalerroroflr = h.apperror(erroroflr, totalerroroflr)
#plot of error hist (only centerx and centery needed)
ylabel = [
"Mittelpunktx Fehler [px]", "Mittelpunkty Fehler [px]",
"Breite Fehler [px]", "Seitenverhätnis Fehler"
]
h.hist(
np.array(errorof), "cen", ylabel,
["norm. Anz. an Fehlern"] * 4,
clippath + "/analysis/pre/" + clipname + "ofhist" + "win" +
str(windows[k]))
ylabel = [
"linksobenx [px]", "linksobeny [px]", "rechtsobenx [px]",
"rechtsobeny [px]"
]
h.hist(
np.array(erroroflr), "cen", ylabel,
["norm. Anz. an Fehlern"] * 4,
clippath + "/analysis/pre/" + clipname + "ofhist" + "win" +
str(windows[k]))
#plot overall errors as histogramms (over all clips)
ylabel = [
"Mittelpunktx Fehler [px]", "Mittelpunkty Fehler [px]",
"Breite Fehler [px]", "Seitenverhätnis Fehler"
]
h.hist(np.array(totalerrorofcen), "cen", ylabel,
["norm. Anz. an Fehlern"] * 4,
macropath + "pre/preofhist" + "win" + str(windows[k]))
ylabel = [
"linksobenx [px]", "linksobeny [px]", "rechtsobenx [px]",
"rechtsobeny [px]"
]
h.hist(np.array(totalerroroflr), "cen", ylabel,
["norm. Anz. an Fehlern"] * 4,
macropath + "pre/preofhist" + "win" + str(windows[k]))
#save errors to save std and mean from best performing optical flow for kalman filter (the one with highest windowsize in this case)
if k == len(windows) - 1:
totalerrorofcenmem = totalerrorofcen
#ploterrors over all clips
totalerrorcor = []
totalerrorcen = []
totalerrorasp = []
#loop over clips and extract bboxes and gt --> then sum up error
for i in range(np.shape(info)[0]):
clippath = (datapath + mainvid + "/" + info[i][2])
clipname = h.extract(clippath)
#read bboxes
bboxcor = h.readmaarray(clippath + "/detections/clean")[:, 2:6]
gtcor = h.readmaarray(clippath + "/groundtruth/gt")[:, 2:6]
bboxcen = h.corcen(bboxcor)
gtcen = h.corcen(gtcor)
bboxasp = h.corasp(bboxcor)
gtasp = h.corasp(gtcor)
errorcor = h.error(bboxcor, gtcor)
totalerrorcor = h.apperror(errorcor, totalerrorcor)
errorcen = h.error(bboxcen, gtcen)
totalerrorcen = h.apperror(errorcen, totalerrorcen)
errorasp = h.error(bboxasp, gtasp)
totalerrorasp = h.apperror(errorasp, totalerrorasp)
h.hist(np.array(totalerrorcor), "cor", [
"xmin Fehler [px]", "ymin Fehler [px]", "xmax Fehler [px]",
"ymax Fehler [px]"
], ["norm. Anz. an Fehlern"] * 4, macropath + "/pre/prehist")
h.hist(np.array(totalerrorcen), "cen", [
"Mittelpunktx Fehler [px]", "Mittelpunkty Fehler [px]",
"Breite Fehler [px]", "Höhe Fehler [px]"
], ["norm. Anz. an Fehlern"] * 4, macropath + "pre/prehist")
h.hist(np.array(totalerrorasp), "asp", [
"Mittelpunktx Fehler [px]", "Mittelpunkty Fehler [px]",
"Breite Fehler [px]", "Seitenverhältnis Fehler"
], ["norm. Anz. an Fehlern"] * 4, macropath + "pre/prehist")
#save mean and std as txt
meancor, stdcor, n = h.stats(totalerrorcor)
meancen, stdcen, _ = h.stats(totalerrorcen)
meanasp, stdasp, _ = h.stats(totalerrorasp)
meanof, stdof, nof = h.stats(totalerrorofcenmem)
np.savetxt(
macropath + "pre/prestatscor.txt", (meancor, stdcor),
fmt="%1.2f",
header=
"mean(first row) and std(second row) for (xmin,ymin,xmax,ymax)" +
"datapoints: " + str(n))
np.savetxt(
macropath + "pre/prestatscen.txt", (meancen, stdcen),
fmt="%1.2f",
header=
"mean(first row) and std(second row) for (centerx,centery,width,height)"
+ "datapoints: " + str(n))
np.savetxt(
macropath + "pre/prestatsasp.txt", (meanasp, stdasp),
fmt="%1.2f",
header=
"mean(first row) and std(second row) for (centerx,centery,width,Seitenverhältnis)"
+ "datapoints: " + str(n))
np.savetxt(
macropath + "pre/prestatsof.txt", (meanof, stdof),
fmt="%1.2f",
header=
"mean(first row) and std(second row) for optical flow init with gt (compared to gt center and width)"
+ "datapoints: " + str(nof))
print("error pre analysis done and histogramms saved")
#create array for mean and std to give kalman function 4 x 2 x 4
statsinfo = np.zeros((4, 2, 4))
statsinfo[0, 0, :] = meancor
statsinfo[1, 0, :] = meancen
statsinfo[2, 0, :] = meanasp
statsinfo[3, 0, :3] = meanof
statsinfo[0, 1, :] = stdcor
statsinfo[1, 1, :] = stdcen
statsinfo[2, 1, :] = stdasp
statsinfo[3, 1, :3] = stdof
#save it to file to use it later
np.save(macropath + "pre/statsdata.npy", statsinfo)
#deletetextfile
h.deletefile(macropath + "pre/preavgioucorroverview.txt")
#loop over clips and save iou of yolo/gt as txt
ioumem = []
for i in range(np.shape(info)[0]):
clippath = (datapath + mainvid + "/" + info[i][2])
clipname = h.extract(clippath)
bboxclean = h.readmaarray(clippath + "/detections/clean")[:, 2:6]
gt = h.readmaarray(clippath + "/groundtruth/gt")[:, 2:6]
iou = h.iouclip(bboxclean, gt)
np.savetxt(
clippath + "/analysis/pre/preavgiou.txt",
[np.mean(iou), np.std(iou)],
fmt="%1.2f",
header="avg. iou of whole clip, mean(1strow) + std(2nd row)")
h.writeapp(
macropath + "pre/preavgiouoverview.txt", clipname + ": mean " +
str(np.mean(iou)) + " std " + str(np.std(iou)) + "\n")
np.savetxt(clippath + "/analysis/pre/preiou.txt",
iou,
fmt="%1.2f",
header="iou of whole clip")
ioumem.extend(iou)
plt.hist(ioumem)
plt.tight_layout()
plt.savefig(macropath + "pre/preiouspread.pdf")
plt.close("all")
#save avg iou and avg iou as txt
np.savetxt(
macropath + "pre/preiou.txt",
ioumem,
fmt="%1.2f",
header=
"iou of all clips(compared at places where yolo originally detected a bbox)"
)
np.savetxt(
macropath + "pre/preavgiou.txt",
[np.mean(ioumem), np.std(ioumem)],
fmt="%1.2f",
header=
"avgiou of all clips, mean + std (compared at places where yolo originally detected a bbox)"
)
h.writeapp(
macropath + "pre/preavgiouoverview.txt", "total" + ": mean " +
str(np.mean(ioumem)) + " std " + str(np.std(ioumem)) + "\n")
print("IoU pre analysis done")
print("pre-analysis done")
"""
####################################################################
correct errors
####################################################################
"""
print("correcting errors")
statsinfo = np.load(macropath + "pre/statsdata.npy")
for i in range(np.shape(info)[0]):
#define clippath to give different methods
clippath = (datapath + mainvid + "/" + info[i][2])
clipname = h.extract(clippath)
bboxclean = h.readmaarray(clippath + "/detections/clean")[:, 2:6]
bboxcleancorr = h.correctmean(bboxclean, statsinfo[0, 0, :])
h.writemaarray(bboxcleancorr, clippath + "/detections/cleancorr",
"cleaned up detections + corrected with mean")
### plot histograms again (with corrected bbox)
totalerrorcor = []
totalerrorcen = []
totalerrorasp = []
#loop over clips and extract bboxes and gt --> then sum up error
for i in range(np.shape(info)[0]):
clippath = datapath + mainvid + "/" + info[i][2]
clipname = h.extract(clippath)
#read bboxes
bboxcor = h.readmaarray(clippath + "/detections/cleancorr")
gtcor = h.readmaarray(clippath + "/groundtruth/gt")[:, 2:6]
bboxcen = h.corcen(bboxcor)
gtcen = h.corcen(gtcor)
bboxasp = h.corasp(bboxcor)
gtasp = h.corasp(gtcor)
errorcor = h.error(bboxcor, gtcor)
totalerrorcor = h.apperror(errorcor, totalerrorcor)
errorcen = h.error(bboxcen, gtcen)
totalerrorcen = h.apperror(errorcen, totalerrorcen)
errorasp = h.error(bboxasp, gtasp)
totalerrorasp = h.apperror(errorasp, totalerrorasp)
h.hist(np.array(totalerrorcor), "cor", [
"xmin Fehler [px]", "ymin Fehler [px]", "xmax Fehler [px]",
"ymax Fehler [px]"
], ["norm. Anz. an Fehlern"] * 4, macropath + "/pre/prehistcorr")
h.hist(np.array(totalerrorcen), "cen", [
"Mittelpunktx Fehler [px]", "Mittelpunkty Fehler [px]",
"Breite Fehler [px]", "Höhe Fehler [px]"
], ["norm. Anz. an Fehlern"] * 4, macropath + "pre/prehistcorr")
h.hist(np.array(totalerrorasp), "asp", [
"Mittelpunktx Fehler [px]", "Mittelpunkty Fehler [px]",
"Breite Fehler [px]", "Seitenverhältnis Fehler"
], ["norm. Anz. an Fehlern"] * 4, macropath + "pre/prehistcorr")
#deletetextfile
h.deletefile(macropath + "pre/preavgioucorroverview.txt")
#loop to get iou
ioumem = []
for i in range(np.shape(info)[0]):
clippath = datapath + mainvid + "/" + info[i][2]
clipname = h.extract(clippath)
bboxcleancorr = h.readmaarray(clippath + "/detections/cleancorr")
gt = h.readmaarray(clippath + "/groundtruth/gt")[:, 2:6]
iou = h.iouclip(bboxcleancorr, gt)
np.savetxt(
clippath + "/analysis/pre/preavgioucorr.txt",
[np.mean(iou), np.std(iou)],
fmt="%1.2f",
header="avg. iou of whole clip, mean(1strow) + std(2nd row)")
h.writeapp(
macropath + "pre/preavgioucorroverview.txt",
clipname + ": mean " + str(np.mean(iou)) + " std " +
str(np.std(iou)) + "\n")
np.savetxt(clippath + "/analysis/pre/preioucorr.txt",
iou,
fmt="%1.2f",
header="iou of whole clip")
ioumem.extend(iou)
plt.hist(ioumem)
plt.tight_layout()
plt.savefig(macropath + "pre/preiouspreadcorr.pdf")
plt.close("all")
#save avg iou and avg iou as txt
np.savetxt(
macropath + "pre/preioucorr.txt",
ioumem,
fmt="%1.2f",
header=
"iou of all clips(compared at places where yolo originally detected a bbox)"
)
np.savetxt(
macropath + "pre/preavgioucorr.txt",
[np.mean(ioumem), np.std(ioumem)],
fmt="%1.2f",
header=
"avgiou of all clips, mean + std (compared at places where yolo originally detected a bbox)"
)
h.writeapp(
macropath + "pre/preavgioucorroverview.txt", "total" + ": mean " +
str(np.mean(ioumem)) + " std " + str(np.std(ioumem)) + "\n")
print("correcting errors done")
"""
####################################################################
algorithms
####################################################################
"""
"""
####################################################################
main kalman filter loops
####################################################################
"""
#define different models for kalman filter (pair model with representation --> need representation to be able to know how to transform bboxes)
modelinfo = np.array(
["simplecen", "cenof", "aspof", "aspofwidth", "aspofwidthman"])
if skip < 4:
statsinfo = np.load(macropath + "pre/statsdata.npy")
#create a bunch of folders for the different models
for i in range(np.shape(info)[0]):
#define clippath to give different methods
clippath = datapath + mainvid + "/" + info[i][2]
clipname = h.extract(clippath)
for j in range(modelinfo.shape[0]):
h.makedirsx(clippath + "/algorithms/kalman/" + modelinfo[j])
h.makedirsx(clippath + "/analysis/post/" + modelinfo[j])
h.makedirsx(macropath + "post/" + modelinfo[j])
#use kalman filter on all clips repr: aspect ratio
#TODO have custom line here with representations(can be 2 models for 1 representation)
#loop over those
for i in range(np.shape(info)[0]):
clippath = datapath + mainvid + "/" + info[i][2]
clipname = h.extract(clippath)
bboxcleancorr = h.readmaarray(clippath + "/detections/cleancorr")
gt = h.readmaarray(clippath + "/groundtruth/gt")[:, 2:6]
ofpoints = h.readmaarray(clippath + "/algorithms/of/" + clipname +
"ofcenwidth" + "win" + str(windows[2]))
for j in range(modelinfo.shape[0]):
#transform depending on what representation is used:
result, memp, memk = kal.kalclip(bboxcleancorr,
gt,
modelinfo[j],
clippath,
statsinfo,
ofpoint=ofpoints)
h.writemaarray(
result, clippath + "/algorithms/kalman/" + modelinfo[j] +
"/" + modelinfo[j], "kalman filter results")
h.viskal(
bboxcleancorr, gt, result, clippath, clippath +
"/algorithms/kalman/" + modelinfo[j] + "/" + modelinfo[j])
end = h.lastindex(gt)
#plot p matrix and k matrix over time
h.plotp(
memp, end, clippath + "/algorithms/kalman/" +
modelinfo[j] + "/" + clipname + modelinfo[j])
h.plotk(
memk, end, clippath + "/algorithms/kalman/" +
modelinfo[j] + "/" + clipname + modelinfo[j])
print("kalman filter ran on:" + clipname)
print("kalman filter used on clips and results saved")
"""
#########################################################
post analysis
#########################################################
"""
if skip < 5:
#analysis after using algorithm
#plots over time for all representations and the results of all models each with all different parameters
for i in range(np.shape(info)[0]):
clippath = datapath + mainvid + "/" + info[i][2]
clipname = h.extract(clippath)
for j in range(modelinfo.shape[0]):
results = h.readmaarray(clippath + "/algorithms/kalman/" +
modelinfo[j] + "/" + modelinfo[j])
gt = h.readmaarray(clippath + "/groundtruth/gt")[:, 2:6]
bboxcleancorr = h.readmaarray(clippath +
"/detections/cleancorr")
#corner representation
#plot and save results
ylabel = ["xmin [px]", "ymin [px]", "xmax [px]", "ymax [px]"]
h.timeplot3(
gt, bboxcleancorr, results, ylabel, ["Frame [k]"] * 4,
clippath + "/analysis/post/" + modelinfo[j] + "/" +
clipname + modelinfo[j] + "otkal")
#center representation
#plot and save results
ylabel = [
"Mittelpunktx [px]", "Mittelpunkty [px]", "Breite [px]",
"Höhe [px]"
]
h.timeplot3(
h.corcen(gt), h.corcen(bboxcleancorr), h.corcen(results),
ylabel, ["Frame [k]"] * 4, clippath + "/analysis/post/" +
modelinfo[j] + "/" + clipname + modelinfo[j] + "otkal")
#aspect ratio representation
#plot and save yolo bboxes
ylabel = [
"Mittelpunktx [px]", "Mittelpunkty [px]", "Breite [px]",
"Seitenverhältnis"
]
h.timeplot3(
h.corasp(gt), h.corasp(bboxcleancorr), h.corasp(results),
ylabel, ["Frame [k]"] * 4, clippath + "/analysis/post/" +
modelinfo[j] + "/" + clipname + modelinfo[j] + "otkal")
print("post analysis time plot on:" + clipname)
#plot errors over time
for i in range(np.shape(info)[0]):
clippath = datapath + mainvid + "/" + info[i][2]
clipname = h.extract(clippath)
for j in range(modelinfo.shape[0]):
results = h.readmaarray(clippath + "/algorithms/kalman/" +
modelinfo[j] + "/" + modelinfo[j])
gt = h.readmaarray(clippath + "/groundtruth/gt")[:, 2:6]
#corner representation
#plot error and save results
ylabel = [
"xmin Fehler [px]", "ymin Fehler [px]", "xmax Fehler [px]",
"ymax Fehler [px]"
]
h.errorplot(
gt, results, ylabel, ["Frame [k]"] * 4,
clippath + "/analysis/post/" + modelinfo[j] + "/" +
clipname + modelinfo[j] + "otkalerror")
#plot rmse
ylabel = [
"xmin RMSE [px]", "ymin RMSE [px]", "xmax RMSE [px]",
"ymax RMSE [px]"
]
h.rmseplot(
gt, results, ylabel, ["Frame [k]"] * 4,
clippath + "/analysis/post/" + modelinfo[j] + "/" +
clipname + modelinfo[j] + "otkalrmse")
#save rmse as txt
np.savetxt(clippath + "/analysis/post/" + modelinfo[j] + "/" +
clipname + modelinfo[j] + "rmsecor.txt",
(h.calcrmse(gt, results)),
fmt="%1.2f",
header="rmse for xmin,ymin,xmax,ymax")
#center representation
#plot and save results
ylabel = [
"Mittelpunktx Fehler [px]", "Mittelpunkty Fehler [px]",
"Breite Fehler [px]", "Höhe Fehler [px]"
]
h.errorplot(
h.corcen(gt), h.corcen(results), ylabel, ["Frame [k]"] * 4,
clippath + "/analysis/post/" + modelinfo[j] + "/" +
clipname + modelinfo[j] + "otkalerror")
ylabel = [
"Mittelpunktx RMSE [px]", "Mittelpunkty RMSE [px]",
"Breite RMSE [px]", "Höhe RMSE [px]"
]
h.rmseplot(
h.corcen(gt), h.corcen(results), ylabel, ["Frame [k]"] * 4,
clippath + "/analysis/post/" + modelinfo[j] + "/" +
clipname + modelinfo[j] + "otkalrmse")
#save rmse as txt
np.savetxt(clippath + "/analysis/post/" + modelinfo[j] + "/" +
clipname + modelinfo[j] + "rmsecen.txt",
(h.calcrmse(h.corcen(gt), h.corcen(results))),
fmt="%1.2f",
header="rmse for centerx,centery,width,height")
#aspect ratio representation
#plot and save yolo bboxes
ylabel = [
"Mittelpunktx Fehler [px]", "Mittelpunkty Fehler [px]",
"Breite Fehler [px]", "Seitenverhältnis Fehler"
]
h.errorplot(
h.corasp(gt), h.corasp(results), ylabel, ["Frame [k]"] * 4,
clippath + "/analysis/post/" + modelinfo[j] + "/" +
clipname + modelinfo[j] + "otkalerror")
#plot rmse
ylabel = [
"Mittelpunktx RMSE [px]", "Mittelpunkty RMSE [px]",
"Breite RMSE [px]", "Seitenverhältnis RMSE"
]
h.rmseplot(
h.corasp(gt), h.corasp(results), ylabel, ["Frame [k]"] * 4,
clippath + "/analysis/post/" + modelinfo[j] + "/" +
clipname + modelinfo[j] + "otkalrmse")
#save rmse as txt
np.savetxt(clippath + "/analysis/post/" + modelinfo[j] + "/" +
clipname + modelinfo[j] + "rmseasp.txt",
(h.calcrmse(h.corasp(gt), h.corasp(results))),
fmt="%1.2f",
header="rmse for centerx,centery,width,asp")
print("post analysis rmse time plot on:" + clipname)
#plot errors after algorithms
for j in range(modelinfo.shape[0]):
totalerrorcor = []
totalerrorcen = []
totalerrorasp = []
#loop over clips and extract bboxes and gt --> then sum up error
for i in range(np.shape(info)[0]):
clippath = (datapath + mainvid + "/" + info[i][2])
clipname = h.extract(clippath)
#read bboxes
resultscor = h.readmaarray(clippath + "/algorithms/kalman/" +
modelinfo[j] + "/" + modelinfo[j])
gtcor = h.readmaarray(clippath + "/groundtruth/gt")[:, 2:6]
resultscen = h.corcen(resultscor)
gtcen = h.corcen(gtcor)
resultsasp = h.corasp(resultscor)
gtasp = h.corasp(gtcor)
errorcor = h.error(resultscor, gtcor)
totalerrorcor = h.apperror(errorcor, totalerrorcor)
errorcen = h.error(resultscen, gtcen)
totalerrorcen = h.apperror(errorcen, totalerrorcen)
errorasp = h.error(resultsasp, gtasp)
totalerrorasp = h.apperror(errorasp, totalerrorasp)
h.hist(
np.array(totalerrorcor), "cor", [
"xmin Fehler [px]", "ymin Fehler [px]", "xmax Fehler [px]",
"ymax Fehler [px]"
], ["norm. Anz. an Fehlern"] * 4, macropath + "post/" +
modelinfo[j] + "/" + modelinfo[j] + "posthist")
h.hist(
np.array(totalerrorcen), "cen", [
"Mittelpunktx Fehler [px]", "Mittelpunkty Fehler [px]",
"Breite Fehler [px]", "Höhe Fehler [px]"
], ["norm. Anz. an Fehlern"] * 4, macropath + "post/" +
modelinfo[j] + "/" + modelinfo[j] + "posthist")
h.hist(
np.array(totalerrorasp), "asp", [
"Mittelpunktx Fehler [px]", "Mittelpunkty Fehler [px]",
"Breite Fehler [px]", "Seitenverhältnis Fehler"
], ["norm. Anz. an Fehlern"] * 4, macropath + "post/" +
modelinfo[j] + "/" + modelinfo[j] + "posthist")
#save mean and std as txt
meancor, stdcor, n = h.stats(totalerrorcor)
meancen, stdcen, _ = h.stats(totalerrorcen)
meanasp, stdasp, _ = h.stats(totalerrorasp)
np.savetxt(
macropath + "post/" + modelinfo[j] + "/" + modelinfo[j] +
"meanstdcor.txt", (meancor, stdcor),
fmt="%1.2f",
header=
"mean(first row) and std(second row) for (xmin,ymin,xmax,ymax)"
+ "datapoints: " + str(n))
np.savetxt(
macropath + "post/" + modelinfo[j] + "/" + modelinfo[j] +
"meanstdcen.txt", (meancen, stdcen),
fmt="%1.2f",
header=
"mean(first row) and std(second row) for (centerx,centery,width,height)"
+ "datapoints: " + str(n))
np.savetxt(
macropath + "post/" + modelinfo[j] + "/" + modelinfo[j] +
"meanstdasp.txt", (meanasp, stdasp),
fmt="%1.2f",
header=
"mean(first row) and std(second row) for (centerx,centery,width,aspect ratio)"
+ "datapoints: " + str(n))
print("post error analysis done")
#loop over different parameters of algorithms
for j in range(modelinfo.shape[0]):
#deletetextfile
h.deletefile(macropath + "post/" + modelinfo[j] + "/" +
modelinfo[j] + "avgiouoverview.txt")
#loop over clips and save iou of yolo/gt as txt
ioumem = []
for i in range(np.shape(info)[0]):
clippath = (datapath + mainvid + "/" + info[i][2])
clipname = h.extract(clippath)
results = h.readmaarray(clippath + "/algorithms/kalman/" +
modelinfo[j] + "/" + modelinfo[j])
gt = h.readmaarray(clippath + "/groundtruth/gt")[:, 2:6]
bboxcleancorr = h.readmaarray(clippath +
"/detections/cleancorr")[:, 2:6]
h.plotiou(
gt, results, ["IoU"], ["Frame [k]"],
clippath + "/analysis/post/" + modelinfo[j] + "/" +
clipname + modelinfo[j] + "iouot")
iou = h.iouclip(results, gt)
np.savetxt(
clippath + "/analysis/post/" + modelinfo[j] + "/" +
clipname + modelinfo[j] + "avgiou.txt",
[np.mean(iou), np.std(iou)],
fmt="%1.2f",
header=
"avg. iou of whole clip, mean + std (only when detection exists, no \"punishment\" for no detection)"
)
h.writeapp(
macropath + "post/" + modelinfo[j] + "/" + modelinfo[j] +
"avgiouoverview.txt", clipname + ": mean " +
str(np.mean(iou)) + " std " + str(np.std(iou)) + "\n")
np.savetxt(clippath + "/analysis/post/" + modelinfo[j] + "/" +
clipname + modelinfo[j] + "iou.txt",
iou,
fmt="%1.2f",
header="iou of whole clip")
ioumem.extend(iou)
plt.hist(ioumem)
plt.tight_layout()
plt.savefig(macropath + "post/" + modelinfo[j] + "/" +
modelinfo[j] + "iouspread.pdf")
plt.close("all")
np.savetxt(
macropath + "post/" + modelinfo[j] + "/" + modelinfo[j] +
"iou.txt",
ioumem,
fmt="%1.2f",
header=
"iou of all clips(compared at places where yolo originally detected a bbox)"
)
np.savetxt(
macropath + "post/" + modelinfo[j] + "/" + modelinfo[j] +
"avgiou.txt",
[np.mean(ioumem), np.std(ioumem)],
fmt="%1.2f",
header=
"avgiou of all clips, mean + std(compared at places where yolo originally detected a bbox)"
)
h.writeapp(
macropath + "post/" + modelinfo[j] + "/" + modelinfo[j] +
"avgiouoverview.txt", "total" + ": mean " +
str(np.mean(ioumem)) + " std " + str(np.std(ioumem)) + "\n")
print("iou post analysis done")
print("post-analysis done")
