def update(self):
if isinstance(self.line,matplotlib.image.AxesImage):
# image name
try:
if len(self.xdata)==0 and self.dataName: self.xdata=self.imgData[self.dataName] # empty list reference an empty singleton, not the list we want; adjust here
import Image
if self.xdata[current]==None: img=Image.new('RGBA',(1,1),(0,0,0,0))
else: img=Image.open(self.xdata[current])
self.line.set_data(img)
except IndexError: pass
else:
# regular data
import numpy
# current==-1 avoids copy slicing data in the else part
if current==None or current==-1 or afterCurrentAlpha==1:
self.line.set_xdata(self.xdata); self.line.set_ydata(self.ydata)
self.line2.set_xdata([]); self.line2.set_ydata([])
else:
try: # try if we can extend the first part by one so that lines are connected
self.xdata[:current+1]; preCurrEnd=current+1
except IndexError: preCurrEnd=current
preCurrEnd=current+(1 if len(self.xdata)>current else 0)
self.line.set_xdata(self.xdata[:preCurrEnd]); self.line.set_ydata(self.ydata[:preCurrEnd])
self.line2.set_xdata(self.xdata[current:]); self.line2.set_ydata(self.ydata[current:])
try:
x,y=self.xdata[current],self.ydata[current]
except IndexError: x,y=0,0
# this could be written in a nicer way, very likely
try:
pt=numpy.ndarray((2,),buffer=numpy.array([float(x),float(y)]))
if self.scatter:
self.scatter.set_offsets(pt)
# change rotation of the marker (possibly incorrect)
try:
dx,dy=self.xdata[current]-self.xdata[current-1],self.ydata[current]-self.ydata[current-1]
# smoothing from last n values, if possible
# FIXME: does not show arrow at all if less than window values
#try:
# window=10
# dx,dy=[numpy.average(numpy.diff(dta[current-window:current])) for dta in self.xdata,self.ydata]
#except IndexError: pass
# there must be an easier way to find on-screen derivative angle, ask on the matplotlib mailing list
axes=self.line.get_axes()
p=axes.patch; xx,yy=p.get_verts()[:,0],p.get_verts()[:,1]; size=max(xx)-min(xx),max(yy)-min(yy)
aspect=(size[1]/size[0])*(1./axes.get_data_ratio())
angle=math.atan(aspect*dy/dx)
if dx<0: angle-=math.pi
self.scatter.set_transform(matplotlib.transforms.Affine2D().rotate(angle))
except IndexError: pass
if self.annotation:
if math.isnan(x) or math.isnan(y):
if hasattr(self.annotation,'xyann'): self.annotation.xyann=(x,y)
else: self.annotation.xytext=(0,0)
self.annotation.set_text('') # make invisible, place anywhere
else:
#
if hasattr(self.annotation,'xyann'): self.annotation.xyann=(x,y) # newer MPL versions (>=1.4)
else: self.annotation.xyann=(x,y)
self.annotation.set_text(self.annotation.annotateFmt.format(xy=(float(x),float(y))))
except TypeError: pass # this happens at i386 with empty data, saying TypeError: buffer is too small for requested array
def processOneImage(imageFile, imageLabel):
image = Image.open(imageFile, "rb");
buffer = np.array(image);
IMAGE_WIDTH = string.atoi(G_OPTION.getValue("width"));
IMAGE_HEIGHT = string.atoi(G_OPTION.getValue("height"));
IMAGE_CHANNEL = string.atoi(G_OPTION.getValue("channel"));
buffer = buffer.reshape([IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_CHANNEL])
mode = G_OPTION.getValue("mode")
for char in mode:
if(char == 'm'):
#substract
buffer=subtractmean(buffer)
return
else:
#normalize
if(char == 'n'):
return
else:
#whitening
if(char == "w"):
return
else:
if(char == "c"):
return
def calculate(self, imgs): try: img1 = Image.open(str(imgs[0])) img2 = Image.open(str(imgs[1])) if img1.size != img2.size or img1.getbands() != img2.getbands(): return -1 s = 0 for band_index, band in enumerate(img1.getbands()): m1 = np.array([p[band_index] for p in img1.getdata()]).reshape(*img1.size) m2 = np.array([p[band_index] for p in img2.getdata()]).reshape(*img2.size) s += np.sum(np.abs(m1-m2)) s = s/1000000 (filepath1, filename1) = os.path.split(str(imgs[0])) (filepath2, filename2) = os.path.split(str(imgs[1])) label = filename1 + "-" + filename2 return (label, s) except Exception, e: QMessageBox.warning(self, "Messaggio", str(e), QMessageBox.Ok)
def generateMaps(mapFileName, numMaps, start_state, goal_state):
# Inputs:
# mapFileName - string of the file name (including local path)
# numMaps - number of maps to be generated
#
# Outputs:
# generatedMap_cell - cell array containing numMaps+1 maps
# the first element contains the seed map
#
import PIL
import matplotlib.plyplot as plt
map_seed = PIL.open(mapFileName).convert('L')
bridge_ind = numpy.where(map_seed != 0 & map_seed != 1) # locations in the map where a bridge exists
bridge_probs = map_seed[bridge_ind] # probability for those bridges
N = bridge_ind.shape[0] # Number of bridges
# Holds the output for the generated bridges
generatedMap_cell = []
for i in range(0,numMaps):
temp_map = map_seed
rand_vals = np.rand(N,1) # generate a random value for each bridge
bridge_vals = rand_vals < bridge_probs # are the bridges open?
temp_map[bridge_ind] = bridge_vals # create the maps
generatedMap_cell[i] = temp_map # store the maps in the cell array
#plot the maps for testing
plt.figure(i)
plt.imshow(scipy.misc.imresize(temp_map,5, interp = 'nearest'))
(N, M) = map_seed.shape
(x, y) = numpy.meshgrid(numpy.arange(1,N+1), numpy.arange(1,M+1))
map_struct['map_name'] = mapFileName
map_struct['bridge_locations'] = np.array([x(bridge_ind).T, y(bridge_ind).T])
map_struct['bridge_probabilities'] = bridge_probs
map_struct['seed_map'] = map_seed
map_struct['map_samples'] = generatedMap_cell
map_struct['start'] = start_state
map_struct['goal'] = goal_state
def augment(self, img: PIL):
augmented_images = []
img = img.convert('RGB')
augmented_images.append(img) # append original images
if self.augmentation_params.get('flip_left_right'):
augmented_images.append(self.flip_left_right(img))
if self.augmentation_params.get('flip_up_down'):
augmented_images.append(self.flip_up_down(img))
if self.augmentation_params.get('rotate_random_25'):
augmented_images = augmented_images + self.rotate_random_25(img)
if self.augmentation_params.get('blur'):
augmented_images = augmented_images + self.blur(img)
assert len(
augmented_images
) == self.total_augmented_images + 1, "Augment error returning wrong values"
return augmented_images
def savePlotSequence(P,fileBase,stride=1,imgRatio=(5,7),title=None,titleFrames=20,lastFrames=30):
'''Save sequence of plots, each plot corresponding to one line in history. It is especially meant to be used for :obj:`woo.utils.makeVideo`.
:param stride: only consider every stride-th line of history (default creates one frame per each line)
:param title: Create title frame, where lines of title are separated with newlines (``\\n``) and optional subtitle is separated from title by double newline.
:param int titleFrames: Create this number of frames with title (by repeating its filename), determines how long the title will stand in the movie.
:param int lastFrames: Repeat the last frame this number of times, so that the movie does not end abruptly.
:return: List of filenames with consecutive frames.
'''
data,imgData,plots=P.data,P.imgData,P.plots
fig=createPlots(P,noShow=True,replace=True,subPlots=True,scatterSize=60,wider=True)[0]
sqrtFigs=math.sqrt(len(plots))
fig.set_size_inches(8*sqrtFigs,5*sqrtFigs) # better readable
fig.subplots_adjust(left=.05,right=.95,bottom=.05,top=.95) # make it more compact
if len(plots)==1 and plots[list(plots.keys())[0]]==None: # only pure snapshot is there
fig.set_size_inches(5,5)
fig.subplots_adjust(left=0,right=1,bottom=0,top=1)
#if not data.keys(): raise ValueError("plot.data is empty.")
pltLen=max(len(data[list(data.keys())[0]]) if data else 0,len(imgData[list(imgData.keys())[0]]) if imgData else 0)
if pltLen==0: raise ValueError("Both plot.data and plot.imgData are empty.")
global current
ret=[]
print('Saving %d plot frames, it can take a while...'%(pltLen))
for i,n in enumerate(range(0,pltLen,stride)):
current=n
for l in P.currLineRefs: l.update()
out=fileBase+'-%03d.png'%i
fig.savefig(out)
ret.append(out)
sys.stderr.write('[%d]'%i)
if len(ret)==0: raise RuntimeError("No images created?!")
if title:
import Image
titleImgName=fileBase+'-title.png'
createTitleFrame(titleImgName,Image.open(ret[-1]).size,title)
ret=titleFrames*[titleImgName]+ret
if lastFrames>1: ret+=(lastFrames-1)*[ret[-1]]
return ret
def savePlotSequence(P,fileBase,stride=1,imgRatio=(5,7),title=None,titleFrames=20,lastFrames=30):
'''Save sequence of plots, each plot corresponding to one line in history. It is especially meant to be used for :obj:`woo.utils.makeVideo`.
:param stride: only consider every stride-th line of history (default creates one frame per each line)
:param title: Create title frame, where lines of title are separated with newlines (``\\n``) and optional subtitle is separated from title by double newline.
:param int titleFrames: Create this number of frames with title (by repeating its filename), determines how long the title will stand in the movie.
:param int lastFrames: Repeat the last frame this number of times, so that the movie does not end abruptly.
:return: List of filenames with consecutive frames.
'''
data,imgData,plots=P.data,P.imgData,P.plots
fig=createPlots(P,noShow=True,replace=True,subPlots=True,scatterSize=60,wider=True)[0]
sqrtFigs=math.sqrt(len(plots))
fig.set_size_inches(8*sqrtFigs,5*sqrtFigs) # better readable
fig.subplots_adjust(left=.05,right=.95,bottom=.05,top=.95) # make it more compact
if len(plots)==1 and plots[plots.keys()[0]]==None: # only pure snapshot is there
fig.set_size_inches(5,5)
fig.subplots_adjust(left=0,right=1,bottom=0,top=1)
#if not data.keys(): raise ValueError("plot.data is empty.")
pltLen=max(len(data[data.keys()[0]]) if data else 0,len(imgData[imgData.keys()[0]]) if imgData else 0)
if pltLen==0: raise ValueError("Both plot.data and plot.imgData are empty.")
global current
ret=[]
print 'Saving %d plot frames, it can take a while...'%(pltLen)
for i,n in enumerate(range(0,pltLen,stride)):
current=n
for l in P.currLineRefs: l.update()
out=fileBase+'-%03d.png'%i
fig.savefig(out)
ret.append(out)
sys.stderr.write('[%d]'%i)
if len(ret)==0: raise RuntimeError("No images created?!")
if title:
import Image
titleImgName=fileBase+'-title.png'
createTitleFrame(titleImgName,Image.open(ret[-1]).size,title)
ret=titleFrames*[titleImgName]+ret
if lastFrames>1: ret+=(lastFrames-1)*[ret[-1]]
return ret
def user_color(original, new):
'''allows user to change border colors from set options'''
img = Image.open(original) #opens original image
#adds border to original
color = raw_input('What color would you like the border to be?(Prompt \
will appear until all images are have a chosen color.): \n')
time.sleep(1)
thic = raw_input('What width would you like the border to be?(Prompt \
will appear until all images are have a chosen width.): \n')
try:
img_with_border = ImageOps.expand(img,
border=int(thic),
fill=str(color))
except ValueError:
img_with_border = ImageOps.expand(img, border=30, fill=str("blue"))
try:
img_with_border = ImageOps.expand(img,
border=int(thic),
fill=str(color))
except TypeError:
img_with_border = ImageOps.expand(img, border=30, fill=str("blue"))
return img_with_border
def rescale(data, width, height, force=True):
"""Rescale the given image, optionally cropping it to make sure the result image has the specified width and height."""
max_width = width
max_height = height
input_file = StringIO(data)
img = pil.open(input_file)
if not force:
img.thumbnail((max_width, max_height), pil.ANTIALIAS)
else:
src_width, src_height = img.size
src_ratio = float(src_width) / float(src_height)
dst_width, dst_height = max_width, max_height
dst_ratio = float(dst_width) / float(dst_height)
if dst_ratio < src_ratio:
crop_height = src_height
crop_width = crop_height * dst_ratio
x_offset = float(src_width - crop_width) / 2
y_offset = 0
else:
crop_width = src_width
crop_height = crop_width / dst_ratio
x_offset = 0
y_offset = float(src_height - crop_height) / 3
img = img.crop((x_offset, y_offset, x_offset+int(crop_width), y_offset+int(crop_height)))
img = img.resize((dst_width, dst_height), pil.ANTIALIAS)
tmp = StringIO()
img.save(tmp, 'JPEG')
tmp.seek(0)
output_data = tmp.getvalue()
input_file.close()
tmp.close()
return output_data
def createPlots(P,subPlots=True,noShow=False,replace=True,scatterSize=60,wider=False):
'''Create plots based on current data;
:param subPlots: show all plots in one figure as subplots; otherwise, create multiple figures
:param noShow: use headless backend for plots, and do not show plots on the screen
:param replace: do not close existing figures, and do not update P.currLineRefs
'''
import logging
data,imgData,plots,labels,xylabels,legendLoc,axesWd,annotateFmt=P.data,P.imgData,P.plots,P.labels,P.xylabels,P.legendLoc,P.axesWd,P.annotateFmt
if replace:
if P.currLineRefs:
logging.info('Closing existing figures')
ff=set([l.line.get_axes().get_figure() for l in P.currLineRefs]) # get all current figures
for f in ff: pylab.close(f) # close those
P.currLineRefs=[]
figs=[]
if len(plots)==0: return # nothing to plot
if subPlots:
# compute number of rows and colums for plots we have
subCols=int(round(math.sqrt(len(plots)))); subRows=int(math.ceil(len(plots)*1./subCols))
if wider: subRows,subCols=subCols,subRows
# create a new figure; called once with subPlots, for each subplot without subPlots
def _newFig():
## pylab API
if not noShow: return pylab.figure() # this will go onto the screen; the pylab call sets up the windows as well
else: # with noShow
fig=matplotlib.figure.Figure()
canvas=_HeadlessFigureCanvas(fig) #
return fig
if subPlots: figs=[_newFig()]
for nPlot,p in enumerate(plots.keys()):
pStrip=p.strip().split('=',1)[0]
if not subPlots:
figs.append(_newFig())
axes=figs[-1].add_subplot(1,1,1)
else: axes=figs[-1].add_subplot(subRows,subCols,nPlot+1) # nPlot is 1-based in mpl, for matlab comatibility
axes.grid(True)
if plots[p]==None: # image plot
if not pStrip in imgData.keys(): imgData[pStrip]=[]
# fake (empty) image if no data yet
import Image
if len(imgData[pStrip])==0 or imgData[pStrip][-1]==None: img=Image.new('RGBA',(1,1),(0,0,0,0))
else: img=Image.open(imgData[pStrip][-1])
img=axes.imshow(img,origin='upper')
if replace: P.currLineRefs.append(LineRef(line=img,scatter=None,annotation=None,line2=None,xdata=imgData[pStrip],ydata=None,imgData=imgData,dataName=pStrip))
axes.set_axis_off()
continue
plots_p=[addPointTypeSpecifier(o) for o in tuplifyYAxis(plots[p])]
plots_p_y1,plots_p_y2=[],[]; y1=True
missing=set() # missing data columns
if pStrip not in data.keys(): missing.add(pStrip.decode('utf-8','ignore'))
for d in plots_p:
if d[0]==None:
y1=False; continue
if not isinstance(d[0],(str,unicode)): raise ValueError('Plots specifiers must be strings (not %s)'%(type(d[0]).__name__))
if y1: plots_p_y1.append(d)
else: plots_p_y2.append(d)
try:
if (
d[0] not in data.keys()
# and not callable(d[0])
and not (isinstance(d[0],(str,unicode)) and (d[0].startswith('**') or d[0].startswith('*'))) # hack for callable as strings
# and not hasattr(d[0],'keys')
):
missing.add(d[0])
except UnicodeEncodeError:
import warnings
warnings.error('UnicodeDecodeError when processing data set '+repr(d[0]))
if missing:
if len(data.keys())==0 or len(data[data.keys()[0]])==0: # no data at all yet, do not add garbage NaNs
for m in missing: data[m]=[]
else:
addDataColumns(data,missing)
try:
print 'Missing columns in Scene.plot.data, added NaNs:',', '.join([m.encode('utf-8') for m in missing])
except UnicodeDecodeError:
warnings.warn('UnicodeDecodeError reporting missing data columns -- harmless, just wondering...')
def createLines(pStrip,ySpecs,axes,isY1=True,y2Exists=False):
'''Create data lines from specifications; this code is common for y1 and y2 axes;
it handles y-data specified as callables/dicts passed as string (starting with '*'/'**'), which might create additional lines when updated with liveUpdate.
'''
# save the original specifications; they will be smuggled into the axes object
# the live updated will run yNameFuncs to see if there are new lines to be added
# and will add them if necessary
yNameFuncs=set()
yNames=set()
ySpecs2=[]
for ys in ySpecs:
if not isinstance(ys[0],(str,unicode)): raise ValueError('Plot specifications must be strings (not a %s).'%type(ys[0]))
if ys[0].startswith('**') or ys[0].startswith('*'):
evEx=eval(ys[0][(2 if ys[0].startswith('**') else 1):],{'S':P.scene})
yNameFuncs.add(evEx) # add callable or dictionary
# XXX: what is ys[1]? Previously, there was no line specifier there for dicts at least
# print evEx,type(evEx), evEx.__iter__(),type(evEx.__iter__())
ySpecs2+=[(ret,ys[1]) for ret in evEx] # traverse list or dict keys
else: ySpecs2.append(ys)
if len(ySpecs2)==0:
print 'woo.plot: creating fake plot, since there are no y-data yet'
line,=axes.plot([nan],[nan])
line2,=axes.plot([nan],[nan])
if replace: P.currLineRefs.append(LineRef(line=line,scatter=None,annotation=None,line2=line2,xdata=[nan],ydata=[nan]))
# set different color series for y1 and y2 so that they are recognizable
if matplotlib.rcParams.has_key('axes.color_cycle'): matplotlib.rcParams['axes.color_cycle']='b,g,r,c,m,y,k' if not isY1 else 'm,y,k,b,g,r,c'
for d in ySpecs2:
yNames.add(d)
# should have been handled above already
#if pStrip not in data:
# print 'Missing column %s in Scene.plot.data, added NaN.'%pString
# addDataColumns(data,[pStrip])
if d[0] not in data:
print 'Missing column %s in Scene.plot.data, added NaN.'%d[0]
addDataColumns(data,[d[0]])
line,=axes.plot(data[pStrip],data[d[0]],d[1],label=xlateLabel(d[0],P.labels),**lineKw)
lineKwWithoutAlpha=dict([(k,v) for k,v in lineKw.items() if k!='alpha'])
line2,=axes.plot([],[],d[1],color=line.get_color(),alpha=afterCurrentAlpha,**lineKwWithoutAlpha)
# use (0,0) if there are no data yet
scatterPt=[0,0] if len(data[pStrip])==0 else (data[pStrip][current],data[d[0]][current])
scatterPtPos=[scatterPt[0] if not math.isnan(scatterPt[0]) else 0,scatterPt[1] if not math.isnan(scatterPt[1]) else 0]
# if current value is NaN, use zero instead
scatter=axes.scatter(scatterPtPos[0],scatterPtPos[1],s=scatterSize,color=line.get_color(),**scatterMarkerKw)
if annotateFmt:
if math.isnan(scatterPtPos[0]) or math.isnan(scatterPtPos[1]): text=''
else: text=annotateFmt.format(xy=scatterPt)
annotation=axes.annotate(text,xy=scatterPtPos,color=line.get_color(),**annotateKw)
annotation.annotateFmt=annotateFmt
else: annotation=None
if replace: P.currLineRefs.append(LineRef(line=line,scatter=scatter,annotation=annotation,line2=line2,xdata=data[pStrip],ydata=data[d[0]]))
axes=line.get_axes()
labelLoc=(legendLoc[0 if isY1 else 1] if y2Exists>0 else 'best')
l=axes.legend(loc=labelLoc)
if l:
l.get_frame().set_alpha(legendAlpha)
if hasattr(l,'draggable'): l.draggable(True)
if scientific:
axes.ticklabel_format(style='sci',scilimits=(0,0),axis='both')
# fixes scientific exponent placement for y2: https://sourceforge.net/mailarchive/forum.php?thread_name=20101223174750.GD28779%40ykcyc&forum_name=matplotlib-users
if not isY1: axes.yaxis.set_offset_position('right')
if isY1:
axes.set_ylabel((', '.join([xlateLabel(_p[0],P.labels) for _p in ySpecs2])) if p not in xylabels or not xylabels[p][1] else xylabels[p][1])
axes.set_xlabel(xlateLabel(pStrip,P.labels) if (p not in xylabels or not xylabels[p][0]) else xylabels[p][0])
else:
axes.set_ylabel((', '.join([xlateLabel(_p[0],P.labels) for _p in ySpecs2])) if (p not in xylabels or len(xylabels[p])<3 or not xylabels[p][2]) else xylabels[p][2])
# if there are callable/dict ySpecs, save them inside the axes object, so that the live updater can use those
if yNameFuncs:
axes.wooYNames,axes.wooYFuncs,axes.wooXName,axes.wooLabelLoc=yNames,yNameFuncs,pStrip,labelLoc # prepend woo to avoid clashes
if 0:
# fix missing 'show' method; this has been fixed in matplotlib already, but we need to backport that
# see https://github.com/matplotlib/matplotlib/commit/15fd0ae587a57cb1d7b69546eb359085315148c8
# don't do that for headless backend, error there is fine
fig=axes.get_figure()
if not hasattr(fig,'show'):
mgr=getattr(fig.canvas,'manager')
if mgr: fig.show=lambda *args: mgr.window.show()
createLines(pStrip,plots_p_y1,axes=axes,isY1=True,y2Exists=len(plots_p_y2)>0)
if axesWd>0:
axes.axhline(linewidth=axesWd,color='k')
axes.axvline(linewidth=axesWd,color='k')
# create y2 lines, if any
if len(plots_p_y2)>0:
axes=axes.twinx() # create the y2 axis
createLines(pStrip,plots_p_y2,axes,isY1=False,y2Exists=True)
### scene is not directly accessible from here, do it like this:
S=woo.master.scene
if S.plot==P:
if 'title' in S.tags: axes.set_title(S.tags['title'])
return figs
#Play Video
import cv2
import mediapipe as mp
import streamlit as st
mp_drawing = mp.solutions.drawing_utils
mp_hands = mp.solutions.hands
y=["image"]
temp_status = len(os.listdir("image"))
#img = None
uploaded_file = st.file_uploader("Alternatively upload a file")
if uploaded_file is not None:
img = Image.open(uploaded_file)
img.save("image/hands.jpg")
st.success("Uploaded.")
# For static images:
hands = mp_hands.Hands(
static_image_mode=True,
max_num_hands=2,
min_detection_confidence=0.5)
st.write("Raw image")
try:
st.image(img, use_column_width=True)
except NameError:
pass
def __init__(self, dataroot, train=True, augment=True):
self.images = []
self.bubbles = []
self.labels = []
dataset = [] # contains tuples of images and associated 360 bubbles
if train:
file = np.loadtxt(os.path.join(dataroot, "dataset_train.txt"),
dtype=str,
skiprows=3)
else:
file = np.loadtxt(os.path.join(dataroot, "dataset_test.txt"),
dtype=str,
skiprows=3)
# load image pairs and create training/validation labels
for pair in file:
if 'right' in pair[1]:
self.labels.extend([i for i in range(3, 6)])
elif 'left' in pair[1]:
self.labels.extend([i for i in range(0, 3)])
dataset.extend([pair for i in range(3)])
# calculate dataset length
self.data_len = len(dataset)
# transformations when loading images:
PIL = transforms.ToPILImage()
resize = transforms.Compose([transforms.Resize(300)])
bub_size = transforms.Resize(500)
if augment:
self.image_trans = transforms.Compose([
transforms.RandomCrop(224),
transforms.ColorJitter(brightness=0.5,
contrast=0.5,
saturation=0.5,
hue=0.05),
transforms.ToTensor()
])
self.bubble_trans = transforms.Compose([
transforms.ColorJitter(brightness=0.5,
contrast=0.5,
saturation=0.5,
hue=0.05),
transforms.ToTensor()
])
else:
self.image_trans = transforms.Compose(
[transforms.Resize(300),
transforms.ToTensor()])
self.bubble_trans = transforms.Compose([transforms.ToTensor()])
# load images and applying initial preprocessing
for i, (bubble, image) in enumerate(dataset):
image = io.imread(os.path.join(dataroot, image),
plugin='matplotlib')
bubble = io.imread(os.path.join(dataroot, bubble),
plugin='matplotlib')
label = torch.tensor(self.labels[i])
# cropping parameters in height and width, this assumes images of shape (2100, 2800)
params = [1300, 1300]
# set left pixel of the image crop depending on label
if label == 0 or label == 3:
width = 0
elif label == 1 or label == 4:
width = int((image.shape[1] - params[1]) / 2)
elif label == 2 or label == 5:
width = int(image.shape[1] - (params[1] + 1))
# set height of image crop
height = int((image.shape[0] - params[0]) / 2)
# pre-process image files
image = PIL(image)
image = TF.crop(image, height, width, params[0], params[1])
image = resize(image)
self.images.append(image)
# preprocess bubble and add the array
bubble = PIL(bubble)
bubble = bub_size(bubble)
self.bubbles.append(bubble)
def get_rois(
original_img: Image,
PYR_SCALE: float,
WIN_STEP: int,
ROI_SIZE: tuple,
visualize: int,
) -> Tuple[List[np.array], List[Tuple[float, float, float, float]]]:
"""
Descrição
---------
Itera sobre o metodo da piramide para gerar as regiões de interesse
Entradas
--------
original_img (Image)
Imagem de entrada
PYR_SCALE (float)
escala usada no metodo da piramide.
WIN_STEP (int)
passo entre cada window.
ROI_SIZE (float)
tamanho da janela.
visualize (int)
se > 0 entra em modo debug e visualiza cada janela na imagem original
Saidas
------
rois (tuple)
regiões de interesse a serem classificadas
locs (tuple)
localização de cada roi na imagem original
"""
# Inicializa a pirâmide da imagem
pyramid = image_pyramid(original_img, scale=PYR_SCALE, min_size=ROI_SIZE)
# Inicializa a lista de ROIs (Regiões de Interesse) geradas pela
# pirâmide e pela sliding window, e a lista de coordenadas (x, y)
# para guardar a posição de cada ROI na imagem original
rois = []
locs = []
# Tempo inicial do pré-processamento da imagem original
start = time.time()
# Dimensões da imagem original
(_, W) = original_img.shape[:2]
# Aplica o algoritmo da pirâmide sobre a imagem
for image in pyramid:
# Obtém o fator de escala entre a imagem original e
# a camada atual da pirâmide
scale = W / float(image.shape[1])
# Para cada camada da pirâmide, aplica o algoritmo de sliding window
for (x, y, original_roi) in sliding_window(image, WIN_STEP, ROI_SIZE):
# Reescala as coordenadas (x, y) da ROI com respeito às
# dimensões da imagem original
x = int(x * scale)
y = int(y * scale)
w = int(ROI_SIZE[0] * scale)
h = int(ROI_SIZE[1] * scale)
# Transforma a ROI (PIL Image) em array para a classificação
# Atualiza a lista de ROIS e a lista de coordenadas
rois.append(original_roi)
locs.append((x, y, x + w, y + h))
# Checa se as sliding windows devem ser mostradas
if visualize > 0:
# Clona a imagem original e então desenha as bounding boxes
# representando a ROI atual
clone = original_img.copy()
cv2.rectangle(clone, (x, y), (x + w, y + h), (0, 255, 0), 2)
# Mostra a ROI atual na tela
cv2.imshow("Sliding Window", clone)
cv2.imshow("ROI", original_roi)
cv2.waitKey(0)
# Fecha as sliding windows mostradas
if visualize > 0:
cv2.destroyAllWindows()
# Tempo final do pré-processamento da imagem original
end = time.time()
# Duração do pré-processamento da imagem original
print(
f"[INFO] O pré-processamento da imagem durou {end - start:.5f} segundos"
)
return rois, locs
"""
ha, wa = imga.shape[:2]
hb, wb = imgb.shape[:2]
max_height = np.max([ha, hb])
total_width = wa + wb
new_img = np.zeros(shape=(max_height, total_width, 3))
new_img[:ha, :wa] = imga
new_img[:hb, wa:wa + wb] = imgb
return new_img
def concat_n_images(image_path_list):
"""
Combines N color images from a list of image paths.
"""
output = None
for i, img_path in enumerate(image_path_list):
img = plt.imread(img_path)[:, :, :3]
if i == 0:
output = img
else:
output = concat_images(output, img)
return output
for i in glob.glob('D:/GitHub/GitTesis/RAW DATA/*.img'):
fopen = PIL.open(i)
output = concat_n_images(fopen)
plt.imshow(output)
plt.show()
import PIL
img = PIL.Image("1217-Steve-Jobs-Quote-Think-Different.jpg")
win = PIL.ImageWin(img.getWidth(), img.getHeight())
img.draw(win)
img.setDelay(1, 15) # setDelay(0) turns off animation
for row in range(img.getHeight()):
for col in range(img.getWidth()):
p = PIL.getPixel(col, row)
newred = 255 - p.getRed()
newgreen = 255 - p.getGreen()
newblue = 255 - p.getBlue()
newpixel = PIL.Pixel(newred, newgreen, newblue)
img.setPixel(col, row, newpixel)
img.draw(win)
win.exitonclick()
def border_one(original_image, percent_of_side=.3):
""" Rounds the corner of a PIL.Image
original_image must be a PIL.Image
Returns a new PIL.Image with rounded corners, where
0 < percent_of_side < 1 is the corner radius as a portion of the shorter
dimension of original_image
"""
#set the radius of the rounded corners
width, height = original_image.size
###
#create a mask
###
#start with transparent mask
border_mask = PIL.Image.new('RGBA', (width, height), (127, 0, 127, 0))
drawing_layer = PIL.ImageDraw.Draw(border_mask)
# Overwrite the RGBA values with A=255.
# The 127 for RGB values was used merely for visualizing the mask
# Draw two rectangles to fill interior with opaqueness
if width >= 300 or height >= 300:
drawing_layer.polygon([(0, 0), (width, 0), (width, height),
(0, height)],
fill=(127, 0, 127, 170))
drawing_layer.polygon([(0, 0), (30, 0), (30, height), (0, height)],
fill=(0, 0, 255, 0),
outline=(0, 255, 0, 1))
drawing_layer.polygon([(width, 0), (width - 30, 0),
(width - 30, height), (width, height)],
fill=(0, 0, 255, 0),
outline=(0, 255, 0, 1))
drawing_layer.polygon([(0, 0), (width, 0), (width, 30), (0, 30)],
fill=(0, 0, 255, 0),
outline=(0, 255, 0, 1))
drawing_layer.polygon([(0, height), (width, height),
(width, height - 30), (0, height - 30)],
fill=(0, 0, 255, 0),
outline=(0, 255, 0, 1))
# Make the new image, starting with all transparent
result = PIL.Image.new('RGBA', original_image.size, (136, 39, 16, 255))
result.paste(original_image, (0, 0), mask=border_mask)
img = Image.open('ourlogo_test.png')
logo_small = img.resize((85, 93))
result.paste(logo_small, (width - 130, 40), mask=logo_small)
return result
elif width <= 300 and height <= 300:
original_image.size = 500, 500
drawing_layer.polygon([(0, 0), (width, 0), (width, height),
(0, height)],
fill=(127, 0, 127, 170))
drawing_layer.polygon([(0, 0), (20, 0), (20, height), (0, height)],
fill=(0, 0, 255, 0),
outline=(0, 255, 0, 1))
drawing_layer.polygon([(width, 0), (width - 20, 0),
(width - 20, height), (width, height)],
fill=(0, 0, 255, 0),
outline=(0, 255, 0, 1))
drawing_layer.polygon([(0, 0), (width, 0), (width, 20), (0, 20)],
fill=(0, 0, 255, 0),
outline=(0, 255, 0, 1))
drawing_layer.polygon([(0, height), (width, height),
(width, height - 30), (0, height - 30)],
fill=(0, 0, 255, 0),
outline=(0, 255, 0, 1))
# Make the new image, starting with all transparent
result = PIL.Image.new('RGBA', original_image.size, (136, 39, 16, 255))
result.paste(original_image, (0, 0), mask=border_mask)
img = Image.open('ourlogo_test.png')
logo_small = img.resize((10, 23))
result.paste(logo_small, (width - 130, 40), mask=logo_small)
return result
def createPlots(P,subPlots=True,noShow=False,replace=True,scatterSize=60,wider=False):
'''Create plots based on current data;
:param subPlots: show all plots in one figure as subplots; otherwise, create multiple figures
:param noShow: use headless backend for plots, and do not show plots on the screen
:param replace: do not close existing figures, and do not update P.currLineRefs
'''
import logging
data,imgData,plots,labels,xylabels,legendLoc,axesWd,annotateFmt=P.data,P.imgData,P.plots,P.labels,P.xylabels,P.legendLoc,P.axesWd,P.annotateFmt
if replace:
if P.currLineRefs:
logging.info('Closing existing figures')
ff=set([_my_get_axes(l.line).get_figure() for l in P.currLineRefs]) # get all current figures
for f in ff: pylab.close(f) # close those
P.currLineRefs=[]
figs=[]
if len(plots)==0: return # nothing to plot
if subPlots:
# compute number of rows and colums for plots we have
subCols=int(round(math.sqrt(len(plots)))); subRows=int(math.ceil(len(plots)*1./subCols))
if wider: subRows,subCols=subCols,subRows
# create a new figure; called once with subPlots, for each subplot without subPlots
def _newFig():
## pylab API
if not noShow: return pylab.figure() # this will go onto the screen; the pylab call sets up the windows as well
else: # with noShow
fig=matplotlib.figure.Figure()
canvas=_HeadlessFigureCanvas(fig) #
return fig
if subPlots: figs=[_newFig()]
for nPlot,p in enumerate(plots.keys()):
pStrip=p.strip().split('=',1)[0]
if not subPlots:
figs.append(_newFig())
axes=figs[-1].add_subplot(1,1,1)
else: axes=figs[-1].add_subplot(subRows,subCols,nPlot+1) # nPlot is 1-based in mpl, for matlab comatibility
axes.grid(True)
if plots[p]==None: # image plot
if not pStrip in list(imgData.keys()): imgData[pStrip]=[]
# fake (empty) image if no data yet
import Image
if len(imgData[pStrip])==0 or imgData[pStrip][-1]==None: img=Image.new('RGBA',(1,1),(0,0,0,0))
else: img=Image.open(imgData[pStrip][-1])
img=axes.imshow(img,origin='upper')
if replace: P.currLineRefs.append(LineRef(line=img,scatter=None,annotation=None,line2=None,xdata=imgData[pStrip],ydata=None,imgData=imgData,dataName=pStrip))
axes.set_axis_off()
continue
plots_p=[addPointTypeSpecifier(o) for o in tuplifyYAxis(plots[p])]
plots_p_y1,plots_p_y2=[],[]; y1=True
missing=set() # missing data columns
if pStrip not in list(data.keys()): missing.add(pStrip if isinstance(pStrip,str) else pStrip.decode('utf-8','ignore'))
for d in plots_p:
d0=d[0]
if d0==None:
y1=False; continue
if not isinstance(d0,(past.builtins.str,str)): raise ValueError('Plots specifiers must be strings (not %s)'%(type(d[0]).__name__))
if '=' in d0: d0=d0.split('=',1)[0]
if y1: plots_p_y1.append((d0,d[1]))
else: plots_p_y2.append((d0,d[1]))
try:
if (
d0 not in list(data.keys())
# and not callable(d[0])
and not (isinstance(d0,str) and (d0.startswith('**') or d0.startswith('*'))) # hack for callable as strings
# and not hasattr(d[0],'keys')
):
missing.add(d0)
except UnicodeEncodeError:
import warnings
warnings.error('UnicodeDecodeError when processing data set '+repr(d[0]))
if missing:
if len(list(data.keys()))==0 or len(data[list(data.keys())[0]])==0: # no data at all yet, do not add garbage NaNs
for m in missing: data[m]=[]
else:
addDataColumns(data,missing)
try:
print('Missing columns in Scene.plot.data, added NaNs:',', '.join([m for m in missing]))
except UnicodeDecodeError:
warnings.warn('UnicodeDecodeError reporting missing data columns -- harmless, just wondering...')
def createLines(pStrip,ySpecs,axes,isY1=True,y2Exists=False):
'''Create data lines from specifications; this code is common for y1 and y2 axes;
it handles y-data specified as callables/dicts passed as string (starting with '*'/'**'), which might create additional lines when updated with liveUpdate.
'''
# save the original specifications; they will be smuggled into the axes object
# the live updated will run yNameFuncs to see if there are new lines to be added
# and will add them if necessary
yNameFuncs=set()
yNames=set()
ySpecs2=[]
for ys in ySpecs:
if not isinstance(ys[0],(past.builtins.str,str)): raise ValueError('Plot specifications must be strings (not a %s).'%type(ys[0]))
if ys[0].startswith('**') or ys[0].startswith('*'):
evEx=eval(ys[0][(2 if ys[0].startswith('**') else 1):],{'S':P.scene})
yNameFuncs.add(evEx) # add callable or dictionary
# XXX: what is ys[1]? Previously, there was no line specifier there for dicts at least
# print evEx,type(evEx), evEx.__iter__(),type(evEx.__iter__())
ySpecs2+=[(ret,ys[1]) for ret in evEx] # traverse list or dict keys
else: ySpecs2.append(ys)
if len(ySpecs2)==0:
print('woo.plot: creating fake plot, since there are no y-data yet')
line,=axes.plot([nan],[nan])
line2,=axes.plot([nan],[nan])
if replace: P.currLineRefs.append(LineRef(line=line,scatter=None,annotation=None,line2=line2,xdata=[nan],ydata=[nan]))
# set different color series for y1 and y2 so that they are recognizable
if 'axes.color_cycle' in matplotlib.rcParams: matplotlib.rcParams['axes.color_cycle']='b,g,r,c,m,y,k' if not isY1 else 'm,y,k,b,g,r,c'
for d in ySpecs2:
yNames.add(d)
# should have been handled above already
#if pStrip not in data:
# print 'Missing column %s in Scene.plot.data, added NaN.'%pString
# addDataColumns(data,[pStrip])
if d[0] not in data:
print('Missing column %s in Scene.plot.data, added NaN.'%d[0])
addDataColumns(data,[d[0]])
line,=axes.plot(data[pStrip],data[d[0]],d[1],label=xlateLabel(d[0],P.labels),**lineKw)
lineKwWithoutAlpha=dict([(k,v) for k,v in list(lineKw.items()) if k!='alpha'])
line2,=axes.plot([],[],d[1],color=line.get_color(),alpha=afterCurrentAlpha,**lineKwWithoutAlpha)
# use (0,0) if there are no data yet
scatterPt=[0,0] if len(data[pStrip])==0 else (data[pStrip][current],data[d[0]][current])
scatterPtPos=[scatterPt[0] if not math.isnan(scatterPt[0]) else 0,scatterPt[1] if not math.isnan(scatterPt[1]) else 0]
# if current value is NaN, use zero instead
scatter=axes.scatter(scatterPtPos[0],scatterPtPos[1],s=scatterSize,color=line.get_color(),**scatterMarkerKw)
if annotateFmt:
if math.isnan(scatterPtPos[0]) or math.isnan(scatterPtPos[1]): text=''
else: text=annotateFmt.format(xy=scatterPt)
annotation=axes.annotate(text,xy=scatterPtPos,color=line.get_color(),**annotateKw)
annotation.annotateFmt=annotateFmt
else: annotation=None
if replace: P.currLineRefs.append(LineRef(line=line,scatter=scatter,annotation=annotation,line2=line2,xdata=data[pStrip],ydata=data[d[0]]))
axes=_my_get_axes(line)
labelLoc=(legendLoc[0 if isY1 else 1] if y2Exists>0 else 'best')
l=axes.legend(loc=labelLoc)
if l:
l.get_frame().set_alpha(legendAlpha)
if hasattr(l,'draggable'): l.draggable(True)
if scientific:
axes.ticklabel_format(style='sci',scilimits=(0,0),axis='both')
# fixes scientific exponent placement for y2: https://sourceforge.net/mailarchive/forum.php?thread_name=20101223174750.GD28779%40ykcyc&forum_name=matplotlib-users
if not isY1: axes.yaxis.set_offset_position('right')
if isY1:
axes.set_ylabel((', '.join([xlateLabel(_p[0],P.labels) for _p in ySpecs2])) if p not in xylabels or not xylabels[p][1] else xylabels[p][1])
axes.set_xlabel(xlateLabel(pStrip,P.labels) if (p not in xylabels or not xylabels[p][0]) else xylabels[p][0])
else:
axes.set_ylabel((', '.join([xlateLabel(_p[0],P.labels) for _p in ySpecs2])) if (p not in xylabels or len(xylabels[p])<3 or not xylabels[p][2]) else xylabels[p][2])
# if there are callable/dict ySpecs, save them inside the axes object, so that the live updater can use those
if yNameFuncs:
axes.wooYNames,axes.wooYFuncs,axes.wooXName,axes.wooLabelLoc=yNames,yNameFuncs,pStrip,labelLoc # prepend woo to avoid clashes
if 0:
# fix missing 'show' method; this has been fixed in matplotlib already, but we need to backport that
# see https://github.com/matplotlib/matplotlib/commit/15fd0ae587a57cb1d7b69546eb359085315148c8
# don't do that for headless backend, error there is fine
fig=axes.get_figure()
if not hasattr(fig,'show'):
mgr=getattr(fig.canvas,'manager')
if mgr: fig.show=lambda *args: mgr.window.show()
createLines(pStrip,plots_p_y1,axes=axes,isY1=True,y2Exists=len(plots_p_y2)>0)
if axesWd>0:
axes.axhline(linewidth=axesWd,color='k')
axes.axvline(linewidth=axesWd,color='k')
# create y2 lines, if any
if len(plots_p_y2)>0:
axes=axes.twinx() # create the y2 axis
createLines(pStrip,plots_p_y2,axes,isY1=False,y2Exists=True)
### scene is not directly accessible from here, do it like this:
S=woo.master.scene
if S.plot==P:
if 'title' in S.tags: axes.set_title(S.tags['title'])
return figs
def update(self):
if isinstance(self.line,matplotlib.image.AxesImage):
# image name
try:
if len(self.xdata)==0 and self.dataName: self.xdata=self.imgData[self.dataName] # empty list reference an empty singleton, not the list we want; adjust here
import Image
if self.xdata[current]==None: img=Image.new('RGBA',(1,1),(0,0,0,0))
else: img=Image.open(self.xdata[current])
self.line.set_data(img)
except IndexError: pass
else:
# regular data
import numpy
# current==-1 avoids copy slicing data in the else part
if current==None or current==-1 or afterCurrentAlpha==1:
self.line.set_xdata(self.xdata); self.line.set_ydata(self.ydata)
self.line2.set_xdata([]); self.line2.set_ydata([])
else:
try: # try if we can extend the first part by one so that lines are connected
self.xdata[:current+1]; preCurrEnd=current+1
except IndexError: preCurrEnd=current
preCurrEnd=current+(1 if len(self.xdata)>current else 0)
self.line.set_xdata(self.xdata[:preCurrEnd]); self.line.set_ydata(self.ydata[:preCurrEnd])
self.line2.set_xdata(self.xdata[current:]); self.line2.set_ydata(self.ydata[current:])
try:
x,y=self.xdata[current],self.ydata[current]
except IndexError: x,y=0,0
# this could be written in a nicer way, very likely
try:
pt=numpy.ndarray((2,),buffer=numpy.array([float(x),float(y)]))
if self.scatter:
self.scatter.set_offsets(pt)
# change rotation of the marker (possibly incorrect)
try:
dx,dy=self.xdata[current]-self.xdata[current-1],self.ydata[current]-self.ydata[current-1]
# smoothing from last n values, if possible
# FIXME: does not show arrow at all if less than window values
#try:
# window=10
# dx,dy=[numpy.average(numpy.diff(dta[current-window:current])) for dta in self.xdata,self.ydata]
#except IndexError: pass
# there must be an easier way to find on-screen derivative angle, ask on the matplotlib mailing list
# XXX: this can be siplified once we drop support for matplotlib < 1.5
# XXX: and is here to avoid MatplotlibDeprecationWarning for newer versions
axes=_my_get_axes(self.line)
p=axes.patch; xx,yy=p.get_verts()[:,0],p.get_verts()[:,1]; size=max(xx)-min(xx),max(yy)-min(yy)
aspect=(size[1]/size[0])*(1./axes.get_data_ratio())
if dx==0: angle=math.pi
angle=math.atan(aspect*dy/dx)
if dx<0: angle-=math.pi
self.scatter.set_transform(matplotlib.transforms.Affine2D().rotate(angle))
except IndexError: pass
if self.annotation:
if math.isnan(x) or math.isnan(y):
if hasattr(self.annotation,'xyann'): self.annotation.xyann=(x,y)
else: self.annotation.xytext=(0,0)
self.annotation.set_text('') # make invisible, place anywhere
else:
#
if hasattr(self.annotation,'xyann'): self.annotation.xyann=(x,y) # newer MPL versions (>=1.4)
else: self.annotation.xyann=(x,y)
self.annotation.set_text(self.annotation.annotateFmt.format(xy=(float(x),float(y))))
except TypeError: pass # this happens at i386 with empty data, saying TypeError: buffer is too small for requested array
import PIL as IMAGE file_src = 'img processing/imgs/fire/T0t0.jpg' #需要压缩的图片地址 img_src = 'img processing/imgs/fire/yasuo.jpg' #压缩后的图片地址 img = IMAGE.open(file_src) w, h = img.size w, h = round(w * 0.2), round(h * 0.2) # 去掉浮点，防报错 img = img.resize((w, h), IMAGE.ANTIALIAS) img.save(img_src, optimize=True, quality=85) # 质量为85效果最好
plt.figure()
plt.scatter(lidarxx, lidaryy, color=colors, s=0.5)
plt.scatter(edgexx, edgeyy, color='k', s=0.5)
plt.xlim([0, w])
plt.ylim([h, 0])
edgeptsind = lidardepthmapIndrec[edges]
edgepts3d = lidar_camcoord[edgeptsind, :]
edgepts3dprojected = (intrinsic @ edgepts3d.T).T
edgepts3dprojected[:,
0] = edgepts3dprojected[:, 0] / edgepts3dprojected[:, 2]
edgepts3dprojected[:,
1] = edgepts3dprojected[:, 1] / edgepts3dprojected[:, 2]
edgepts3dprojected = edgepts3dprojected[:, 0:3]
plt.figure()
plt.imshow(pil.fromarray(rgbarr))
distances = edgepts3dprojected[:, 2]
colors = cm.jet(1 / distances * 10)
plt.gca().scatter(edgepts3dprojected[:, 0],
edgepts3dprojected[:, 1],
color=colors,
s=0.5)
plt.xlim([0, w])
plt.ylim([h, 0])
from utils import *
edges = (bsmvrec > 0) * (bsmvrec < 0.1)
testsets = [[1610, 558, 1609, 554], [1610, 558, 1606, 558],
[407, 618, 408, 613], [332, 587, 331, 583],
[119, 743, 116, 734], [879, 651, 880, 645],
i += 1
#saving model
torch.save(cn.state_dict(), model_save_path)
print('обучение закончено')
cn = ConvNet(num_classes)
cn.load_state_dict(torch.load(model_save_path))
batch = next(iter(signsValidationLoader))
predictions = cn(batch['image'])
y_test = batch['label']
#print(predictions, y_test)
_, predictions = torch.max(predictions, 1)
plt.imshow(PIL.ToPIL(batch['image'][0]))
print('Gound-true:', dataset.labels[batch['label'][0]])
print('Prediction:', dataset.labels[predictions[0]])
#####Кривые обучения
def smooth_curve(points, factor=0.9):
smoothed_points = []
for point in points:
if smoothed_points:
previous = smoothed_points[-1]
smoothed_points.append(previous * factor + point * (1 - factor))
else:
smoothed_points.append(point)
return smoothed_points
k[int(gt[i][1]), int(gt[i][0])] = 1
# generate density map
k = gaussian_filter_density(k)
# File path to save density map
file_path = img_path.replace('.jpg', '.h5').replace('images', 'ground')
with h5py.File(file_path, 'w') as hf:
hf['density'] = k
# In[10]:
file_path = img_paths[22].replace('.jpg', '.h5').replace('images', 'ground')
print(file_path)
# In[12]:
#Sample Ground Truth
gt_file = h5py.File(file_path, 'r')
groundtruth = np.asarray(gt_file['density'])
plt.imshow(groundtruth, cmap=CM.jet)
print("Sum = ", np.sum(groundtruth))
# In[13]:
# Image corresponding to the ground truth
img = Image.open(file_path.replace('.h5', '.jpg').replace('ground', 'images'))
plt.imshow(img)
print(file_path.replace('.h5', '.jpg').replace('ground', 'images'))
import cyni
import numpy as np
import PIL as Image
cyni.initialize()
device = cyni.getAnyDevice()
device.open()
depthStream = device.createStream("depth", fps=30, width=640, height=480)
#colorStream = device.createStream("color", fps=30, width=1280, height=960)
#colorStream = device.createStream("color", fps=30, width=640, height=480)
#device.setImageRegistrationMode("depth_to_color")
device.setDepthColorSyncEnabled(on=True)
depthStream.start()
# colorStream.start()
# colorFrame = colorStream.readFrame()
# colorFrame = colorStream.readFrame()
# colorFrame = colorStream.readFrame()
# colorFrame = colorStream.readFrame()
depthFrame = depthStream.readFrame()
# registered = cyni.registerColorImage(depthFrame.data, colorFrame.data, depthStream, colorStream)
# Image.fromarray(colorFrame.data).save("color.png")
# Image.fromarray(registered).save("registered.png")
Image.fromarray(cyni.depthMapToImage(depthFrame.data)).save("depth.png")
'''
Python script to crunch sprite assets in Unity by a constant value (for reducing filesize with big images
This automatically applies to .meta files, so nothing changes in game, and it works on Multiple sprite assets as well.
PIL (Python Image Library) needs to be installed to use this.
@author: Brian Intile
'''
import os
from PIL import Image
import PIL
def get_immediate_subdirectories(directory):
return [
name for name in os.listdir(directory)
if os.path.isdir(os.path.join(directory, name))
]
def browse_for_traits(directory, recurse, mult):
for dir_0 in os.listdir(directory):
if (os.path.isfile(directory + u'\\' + dir_0) &
((dir_0.split('.')[-1].lower() == u'png'.lower()) |
(dir_0.split('.')[-1].lower() == u'jpg'.lower()))):
change_value_names(directory + '\\' + dir_0, mult)
if (recurse):
directories = get_immediate_subdirectories(directory)
for x in range(0, len(directories)):
# -*- coding: utf-8 -*- """ Created on Tue Oct 3 17:13:57 2017 @author: cgowl_000 """ import matplotlib as plt import numpy as np import PIL as im r = [[1,0],[0,0]] print(r) g = [[0,1],[0,0]] b = [[0,0],[1,0]] stack = np.dstack((r,b,g)) im.Image(stack) im.
import PIL as Image
pil_im = Image.open('C:\\Users\\Arsalan\\Desktop\\myprogress.PNG')
print(pil_im)
def fetch_card_img_s3(name, bucket=os.environ.get('POKEDEXR_S3')):
s3 = boto3.resource('s3')
x = s3.Object(bucket, f'card_images/original/{name}').get()
src_img = PIL.imread(BytesIO(x['Body'].read()), 0)
return src_img
import PIL import numpy as np img_path = 'data/images_train/data/123033/0e27621e3f8b484c993619abd064948be27f60a2.jpg' img = PIL.imread(img_path) img_array = np.asarray(img) print(img_array.shape)
# -*- coding: utf-8 -*-
"""
Created on Sat Oct 18 12:08:18 2014
@author: girish
"""
import PIL
from scipy.ndimage import filters
from PIL import Image
from pylab import *
import os
im = PIL.Image("banana.jpg")
plot(im)
def filter_detections(
original_img: Image,
preds: List[tuple],
locs: np.array,
min_conf: float,
visualize: int,
) -> Type[None]:
"""
Definição
---------
Mostra as predições
Entradas
--------
original_img (Image)
imagem original.
preds (List[tuple])
lista com as prediçoes (indice, classe, probabilidade).
locs (np.array)
localização das janelas classificadas
min_conf (float)
probabilidade minima para se considerar uma detecçao valida
visualize (int)
variavel de debub, se >0 permite visualizar passos intermediarios do processo
"""
# Inicializa dicionário de predições separado por classes
labels = {}
# Itera sobre todas as predições de cada ROI
for (i, p) in enumerate(preds):
# Separa as informações da ROI atual
(_, label, prob) = p
# Filtra as detecções com baixa prioridade de acordo com min_conf
if prob >= min_conf:
# Seleciona a bounding box da predição
box = locs[i]
# Caso a chave label exista no dicionário labels, guarda seu valor em lab
# Se não, guarda uma lista vazia em lab
lab = labels.get(label, [])
# Adiciona a bounding box e sua probabilidade no dicionário de labels
lab.append((box, prob))
labels[label] = lab
# Itera sobre cada uma das classes detectadas
# Clona a imagem original para desenhar as bounding boxes
clone = original_img.copy()
for label in labels:
# Checa se as bounding boxes encontradas devem ser mostradas
if visualize > 0:
# Clona a imagem original para desenhar as bounding boxes
clone = original_img.copy()
# Itera sobre todas as bounding boxes da classe atual
for (box, _) in labels[label]:
# Desenha as bounding bouxes na imagem
(startX, startY, endX, endY) = box
cv2.rectangle(
clone, (startX, startY), (endX, endY), (0, 255, 0), 2
)
cv2.imshow("Debug", clone)
# Aplica a técnica de non-maxima supression nas predições da classe atual
boxes = np.array([p[0] for p in labels[label]])
proba = np.array([p[1] for p in labels[label]])
boxes = non_max_suppression(boxes, proba)
# Itera sobre as bounding boxes remanescenes
for (startX, startY, endX, endY) in boxes:
# Desenha as bounding bouxes na imagem
cv2.rectangle(
clone, (startX, startY), (endX, endY), (0, 255, 0), 2
)
y = startY - 10 if startY - 10 > 10 else startY + 10
cv2.putText(
clone,
label,
(startX, y),
cv2.FONT_HERSHEY_SIMPLEX,
0.45,
(0, 255, 0),
2,
)
return clone
gt = mat["image_info"][0, 0][0, 0][0]
# Read image
img = plt.imread(img_path)
# Create a zero matrix of image size
k = np.zeros((img.shape[0], img.shape[1]))
# Generate hot encoded matrix of sparse matrix
for i in range(0, len(gt)):
if int(gt[i][1]) < img.shape[0] and int(gt[i][0]) < img.shape[1]:
k[int(gt[i][1]), int(gt[i][0])] = 1
# generate density map
k = gaussian_filter_density(k)
image = Image.fromarray(255 * k)
# image.show()
# File path to save density map
file_path = img_path.replace('.jpg', '.h5').replace('images', 'ground')
with h5py.File(file_path, 'w') as hf:
hf['density'] = k
# %%
path_test = "G:/pycharm/CSRnet-master/data/part_A_final/train_data/ground/IMG_107.h5"
# file_path = img_paths[-5].replace('.jpg','.h5').replace('images','ground')
file_path = path_test
print(file_path)
def imageToNP(imageName):
try:
npa=pil.Image(imageName)
return npa
except(Exception):
print("error")
