def read(self):
from skimage import io
img = io.MultiImage(os.path.expanduser(self.filepath))
if len(img) == 1 and 2 < len(img[0].shape):
img = img[0]
return img
def gen_movie(delta=100,
scaley=0.5,
spacebetween=5,
source_fold=None,
source_file=None,
frame_file=None,
offset=1000,
frame_no=200,
target_fold='/home/mossing/Documents/notebooks/temp/',
frame_rg=(1, 0),
chunk_offset=None):
if chunk_offset is None:
chunk_offset = offset
# source_fold and source_file are format strings, like 'whatever{0}whatever', where {0} specifies where a digit should go
if source_fold is None or source_file is None:
date, animalid, exptno = parse_sbx_filename(frame_file)
source_fold = make_suite2p_fold(date, animalid, exptno)
source_file = make_suite2p_file(date, animalid, exptno)
filenames = [(source_fold + '/' + source_file).format(n)
for n in range(1, 5)]
print('loading images')
print(filenames)
imgs = [skio.MultiImage(filename) for filename in filenames]
print('done loading images')
(Ny, Nx) = imgs[0][0].shape
tr = gen_transform(delta, scaley, (Ny, Nx))
frm_on = gen_frm_on(frame_file, rg=frame_rg)
warped = [skt.warp(imgs[n][chunk_offset], tr) for n in range(4)]
green_frame = np.vstack(
[x[:int(Ny / 2 + spacebetween)] for x in warped[::-1]])
stim_on = gen_stim_on(green_frame.shape)
if not os.path.exists(target_fold):
os.makedirs(target_fold)
print('got here')
for t in range(frame_no):
print(t)
warped = [skt.warp(imgs[n][chunk_offset + t], tr) for n in range(4)]
green_frame = np.vstack(
[x[:int(Ny * scaley + spacebetween)] for x in warped[::-1]])
green_frame = green_frame * (green_frame > 0)
red_frame = np.in1d(offset + t, frm_on) * stim_on
rgb = np.dstack((red_frame, green_frame, red_frame))
plt.imsave(target_fold + '/{0:04d}.tif'.format(t), rgb)
def parse_dir(self, glob_pattern):
from skimage import io
import glob
import os
import numpy as np
files = glob.glob(os.path.join(self.directory, glob_pattern))
array_list = []
for file in files:
multi = io.MultiImage(file, as_gray=True)
im_array = np.stack(multi)
array_list.append(im_array)
im_stack = np.stack(array_list)
return im_stack
def get_slide(slide_name):
"""
Open a whole-slide image (*.tif).
Args:
slide_name: Name of the slide.
Returns:
An skimage object representing a whole-slide image.
"""
# try:
slide = sk.MultiImage(
f'{SRC_TRAIN_DIR}/{slide_name}.{SLIDE_EXT}')[BASE_PAGE]
# except Exception as e:
# print(e.)
return slide
def test_codecs(self):
img = io.MultiImage('/input/tests/data/test.tif')
self.assertEqual((10, 10, 4), img[0].shape)
def ProcessGenerateRecordTask(task):
ident = task['ident']
tiffPath = task['tiffPath']
tfrecordsPath = task['tfrecordsPath']
tileSize = task['tileSize']
outImageSize = task['outImageSize']
rotationStepsCount = task['rotationStepsCount']
#print("processing {0}".format(ident))
#print("reading image from disk {0}".format(ident))
#im = 255 - io.imread(filename,plugin="tifffile")
multiimage = io.MultiImage(tiffPath)
#print("multiimage of {0} elements".format(len(multiimage)))
#for i in range(0,len(multiimage)):
# print("level {0}, shape {1}".format(i,multiimage[i].shape))
im = 255 - multiimage[1]
M = rotationStepsCount
rotStep = 360.0 / M
for i in range(0,M):
tiles = []
tfrecordsPathIdx = "{0}-{1}.tfrecords".format(tfrecordsPath[0:-10], i)
effectiveDegree = rotStep*i
rotated = npImTrans.RotateWithoutCrop(im, effectiveDegree)
#print("getting tiles")
_,tiles = getNotEmptyTiles(rotated, tileSize, emptyCuttOffQuantile=None, emptyCutOffMaxThreshold=10)
if len(tiles) == 0:
print("WARN: Image {0} resulted in 0 tiles. producing blank (black) single tile TfRecords file".format(ident))
tiles = [ np.zeros((outImageSize,outImageSize,3),dtype=np.uint8) ]
else:
# filtering out non green
nonRedTiles = []
for tile in tiles:
red = np.mean(tile[:,:,0]) # supposing RGB, not BGR
green = np.mean(tile[:,:,1])
#print("[R:{0}\tG:{1}; ratio {2}".format(red,green, green/red))
if green / red >= 1.2: # green must be at least 1.5 times more than red (to remove white marker tiles)
nonRedTiles.append(tile)
if len(nonRedTiles)>0:
tiles = nonRedTiles
else:
print("WARN: omited non-green tiles filtering, as the filtering impose the empty tile set")
# normalizing with contrasts
contrasts = []
means = []
#print("normalzing {0} tiles".format(len(tiles)))
for tile in tiles:
mu = npImNorm.getImageMean_withoutBlack(tile, 40.0)
if not np.isnan(mu):
contrast = npImNorm.getImageContrast_withoutPureBlack(tile, precomputedMu=mu)
means.append(mu)
contrasts.append(contrast)
if len(means)>0: # whole image is not entirly black
meanContrast = np.mean(contrasts)
meanMean = np.mean(means)
if meanMean > 0.0:
for j in range(0,len(tiles)):
tiles[j] = npImNorm.GCNtoRGB_uint8(npImNorm.GCN(tiles[j], lambdaTerm=0.0, precomputedContrast=meanContrast, precomputedMean=meanMean), cutoffSigmasRange=2.0)
if outImageSize != tileSize:
resizedTiles = []
for tile in tiles:
resizedTiles.append(cv2.resize(tile, dsize=(outImageSize, outImageSize), interpolation=cv2.INTER_AREA))
tiles = resizedTiles
#print("saving {0}".format(len(tiles)))
savePackAsTFRecord(tiles,tfrecordsPathIdx)
sys.stdout.write("({0}:{1})".format(i,len(tiles)))
sys.stdout.flush()
#print("done")
# debug preview
# N = len(gatheredTiles)
# cols = round(math.sqrt(N))
# rows = math.ceil(N/cols)
# #plt.figure()
# #r,c = tileIdx[N-1]
# #plt.title("tile [{0},{1}]".format(r,c))
# #plt.imshow(gatheredTiles[N-1])
# fig, ax = plt.subplots(rows,cols)
# fig.set_facecolor((0.3,0.3,0.3))
# idx = 1
# for row in range(0,rows):
# for col in range(0,cols):
# row = (idx - 1) // cols
# col = (idx -1) % cols
# #ax[row,col].set_title("tile [{0},{1}]".format(tile_r,tile_c))
# ax[row,col].axis('off')
# if idx-1 < N:
# ax[row,col].imshow(gatheredTiles[idx-1])
# idx = idx + 1
# plt.show() # display it
return len(tiles)
#userinput for file
root = tk.Tk()
root.withdraw()
filename = filedialog.askopenfilename()
filext = os.path.splitext(filename)[1]
#check the file type
if filename.lower().endswith('.avi') or filename.lower().endswith('.mp4'):
filetype = 0
global vid
vid = imageio.get_reader(filename, )
nframes = vid.get_meta_data()['nframes']
elif filename.lower().endswith(('.tiff', '.tif')):
tiffinfo = io.MultiImage(filename)
if len(tiffinfo) > 1:
filetype = 1
nframes = len(tiffinfo)
else:
import glob
filetype = 2
filename = glob.glob(
os.path.split(filename)[0] + os.sep + '*' + filext)
filename.sort()
nframes = len(filename)
elif filename.lower().endswith(('.png', '.jpg', '.jpeg')):
import glob
filetype = 2
filename = glob.glob(os.path.split(filename)[0] + os.sep + '*' + filext)
filename.sort()
def decodeTiff(path):
path = path.decode("utf-8")
image = io.MultiImage(path)[1]
#print("decoded image type {0}. shape {1}. dtype {2}".format(type(image),image.shape, image.dtype))
return image
# Generate an icosphere
# -----------------------------------------------------------------------------
print('Generating icosphere')
icosphere = generate_icosphere(order)
# -----------------------------------------------------------------------------
# Load and process the cubemap image
# -----------------------------------------------------------------------------
print('Loading the cube map data')
# Load the multi-page TIFF image
# Channel 0: RGB
# Channel 1: Depth
# Channel 2: Sematic labels
# Channel 3: Instance labels
tiff = io.MultiImage('inputs/cubemap.tiff')
# Convert RGB image to a torch tensor with dimensions (1, 3, H, W)
cube_rgb = torch.from_numpy(tiff[0]).permute(2, 0, 1).float().unsqueeze(0)
if cuda:
cube_rgb = cube_rgb.cuda()
# Convert depth image to torch tensor with dimensions (1, 1, H, W)
cube_inv_depth = torch.from_numpy(tiff[1].astype(
np.int32)).float().unsqueeze(0).unsqueeze(0)
if cuda:
cube_inv_depth = cube_inv_depth.cuda()
# Convert inverse depth to regular depth
cube_inv_depth[cube_inv_depth == 0] = -1
cube_depth = 1 / cube_inv_depth
# frameId = video.get(1)
# ret, frame = video.read()
# if ret == False:
# break
# if frameId % math.floor(frameRate) == 0:
# cv2.imwrite('images/img'+str(i)+'.jpg', frame)
# i+=1
# video.release()
# cv2.destroyAllWindows()
# ============================
# COLOR QUANTIZATION
# ============================
# === LOAD IMAGE ===
imageCollection = io.MultiImage('./images/*.jpg')
tmp_array = []
# === CONVERT 8 BIT TO FLOAT ===
for image in imageCollection:
converted_image = img_as_float64(image)
# === CONVERT IMAGE INTO 2D MATRIX FOR MANIPULATION ===
image_array = convert_3d_matrix_to_2d(converted_image)
# === TRAIN MODEL TO AGGREGATE COLORS IN ORDER TO HAVE 64 DISTINCT COLORS IN IMAGE ===
image_sample = shuffle(image_array, random_state=0)[:1000]
n_colors = 64
kmeans = KMeans(n_clusters=n_colors, random_state=0).fit(image_sample)
labels = kmeans.predict(image_array)
def analysis(faster_fit,k,II):
import matplotlib.pyplot as plt
import numpy as np
import imageio
import os
if faster_fit:
from edge_detection import linear_subpixel_detection as edge
else:
from edge_detection import errorfunction_subpixel_detection as edge
from skimage.viewer.canvastools import RectangleTool
from skimage.viewer.canvastools import LineTool
from skimage.viewer import ImageViewer
from skimage import io
from shapely.geometry import LineString
import tkinter as tk
from tkinter import filedialog
PO=3 #polyfit order for edge fitting to measure contact angle
thresh=70 #replace with automatic treshold detection at some point
#userinput for file
root = tk.Tk()
root.withdraw()
filename = filedialog.askopenfilename()
filext=os.path.splitext(filename)[1]
#check the file type
if filename.lower().endswith('.avi') or filename.lower().endswith('.mp4'):
filetype=0
global vid
vid = imageio.get_reader(filename,)
nframes = vid.get_meta_data()['nframes']
elif filename.lower().endswith(('.tiff', '.tif')):
tiffinfo=io.MultiImage(filename)
if len(tiffinfo)>1:
filetype=1
nframes=len(tiffinfo)
else:
import glob
filetype=2
filename=glob.glob(os.path.split(filename)[0]+os.sep+'*'+filext)
filename.sort()
nframes=len(filename)
elif filename.lower().endswith(('.png', '.jpg', '.jpeg')):
import glob
filetype=2
filename=glob.glob(os.path.split(filename)[0]+os.sep+'*'+filext)
filename.sort()
nframes=len(filename)
else:
print('unknown filetype')
#function to read a specific frame from the movie, stack, or image sequence
def getcurrframe(filename,framenr,filetype):
def movie():
image=vid.get_data(framenr)[:,:,0]
return image
def tifstack():
stack=io.imread(filename)
image=stack[framenr,:,:]
return image
def images():
image=io.imread(filename[framenr])
return image
filetypes={0 : movie,
1 : tifstack,
2 : images
}
image=filetypes[filetype]()
return image
image = getcurrframe(filename,0,filetype)
#preallocate crop coordinates, maybe unescecarry?
coords=[0,0,0,0]
#show the image and ask for a crop from the user, using skimage.viewer canvastools
viewer = ImageViewer(image)
rect_tool = RectangleTool(viewer, on_enter=viewer.closeEvent) #actually call the imageviewer
viewer.show() #don't forget to show it
coords=np.array(rect_tool.extents)
coords=np.array(np.around(coords),dtype=int)
#crop the image
cropped=image[coords[2]:coords[3],coords[0]:coords[1]]
framesize=cropped.shape
baseinput=np.array([0,0,0,0])
#userinput for the baseline, on which we'll find the contact angles, using skimage.viewer
viewer = ImageViewer(cropped)
line_tool=LineTool(viewer,on_enter=viewer.closeEvent)
viewer.show()
baseinput=line_tool.end_points
#extend the baseline to the edge of the frame (in case the drop grows)
rightbasepoint=np.argmax([baseinput[0,0],baseinput[1,0]])
baseslope=np.float(baseinput[rightbasepoint,1]-baseinput[1-rightbasepoint,1])/(baseinput[rightbasepoint,0]-baseinput[1-rightbasepoint,0])
base=np.array([[0,baseinput[0,1]-baseslope*baseinput[0,0]],[framesize[1],baseslope*framesize[1]+baseinput[0,1]-baseslope*baseinput[0,0]]])
#preallocation of edges, angles and contact points
edgeleft=np.zeros(framesize[0])
edgeright=np.zeros(framesize[0])
thetal=np.zeros(nframes)
thetar=np.zeros(nframes)
contactpointright=np.zeros(nframes)
contactpointleft=np.zeros(nframes)
dropvolume=np.zeros(nframes)
plt.ion()
#loop over frames
for framenr in range (nframes):
image = getcurrframe(filename,framenr,filetype) #get current frame
cropped=np.array(image[round(coords[2]):round(coords[3]),round(coords[0]):round(coords[1])]) #crop frame
edgeleft, edgeright=edge(cropped,thresh) #find the edge with edge function in edge_detection.py
baseline=LineString(base) #using shapely we construct baseline
rightline=LineString(np.column_stack((edgeright,(range(0,framesize[0]))))) #and the lines of the edges
leftline=LineString(np.column_stack((edgeleft,(range(0,framesize[0])))))
leftcontact=baseline.intersection(leftline) #we find the intersectionpoint of the baseline with the edge
rightcontact=baseline.intersection(rightline)
#Detect small drops that are lower than 'k' pixels
#This may break if the drop grows outside the frame on one side. Maybe fix later?
fitpointsleft=edgeleft[range(np.int(np.floor(leftcontact.y)),np.int(np.floor(leftcontact.y)-k),-1)]
if any(fitpointsleft==0):
fitpointsleft=np.delete(fitpointsleft,range(np.argmax(fitpointsleft==0),k))
#polyfit the edges around the baseline, but flipped, because polyfitting a vertical line is bad
leftfit=np.polyfit(range(0,fitpointsleft.shape[0]),fitpointsleft,PO)
leftvec=np.array([1,leftfit[PO-1]]) #vector for angle calculation
fitpointsright=edgeright[range(np.int(np.floor(leftcontact.y)),np.int(np.floor(leftcontact.y)-k),-1)]
if any(fitpointsright==0):
fitpointsright=np.delete(fitpointsright,range(np.argmax(fitpointsright==0),k))
#polyfit the edges around the baseline, but flipped, because polyfitting a vertical line is bad
rightfit=np.polyfit(range(0,fitpointsright.shape[0]),fitpointsright,PO)
rightvec=np.array([1,rightfit[PO-1]]) #vector for angle calculation
basevec=np.array([-baseslope,1]) #base vector for angle calculation (allows for a sloped basline if the camera was tilted)
#calculate the angles using the dot product.
thetal[framenr]=np.arccos(np.dot(basevec,leftvec)/(np.sqrt(np.dot(basevec,basevec))*np.sqrt(np.dot(leftvec,leftvec))))*180/np.pi
thetar[framenr]=180-np.arccos(np.dot(basevec,rightvec)/(np.sqrt(np.dot(basevec,basevec))*np.sqrt(np.dot(rightvec,rightvec))))*180/np.pi
if framenr % II ==0: #plot every II frames, to get a visual indication of how far along the script is
plt.clf()
plt.imshow(image[round(coords[2]):round(coords[3]),round(coords[0]):round(coords[1])],cmap='gray',interpolation="nearest")
plt.plot(edgeleft,range(0,framesize[0]))
plt.plot(edgeright,range(0,framesize[0]))
plt.plot([base[0,0],base[1,0]],[base[0,1],base[1,1]])
plt.title('frame %d of %d' % (framenr, nframes))
plt.pause(0.001)
#save the contact point (the point where the polyfit intersects the baseline)
contactpointright[framenr]=rightfit[PO]
contactpointleft[framenr]=leftfit[PO]
for height in range (0,min(np.int(np.floor(leftcontact.y)),np.int(np.floor(rightcontact.y)))):
dropvolume[framenr]=dropvolume[framenr]+np.pi*np.square((edgeright[height]-edgeleft[height])/2) #volume of each slice in pixels^3, without taking a possible slanted baseline into account
#using cylindrical slice we calculate the remaining volume
slantedbasediff=max(np.floor(leftcontact.y),np.floor(rightcontact.y))-min(np.floor(leftcontact.y),np.floor(rightcontact.y))
#we assume that the radius is constant over the range of the slanted baseline, for small angles this is probably accurate, but for larger angles this is probably wrong.
baseradius=(edgeright[np.int(min(np.floor(leftcontact.y),np.floor(rightcontact.y)))]-edgeleft[np.int(min(np.floor(leftcontact.y),np.floor(rightcontact.y)))])/2
dropvolume[framenr]=dropvolume[framenr]+.5*np.pi*np.square(baseradius)*slantedbasediff
#%%
fitsamplesize=3
if nframes>2*fitsamplesize+1:
leftspeed=np.zeros(nframes)
rightspeed=np.zeros(nframes)
for framenr in range(fitsamplesize,nframes-fitsamplesize-1):
rightposfit=np.polyfit(range(-fitsamplesize,fitsamplesize),contactpointright[range(framenr-fitsamplesize,framenr+fitsamplesize)],1)
leftposfit=np.polyfit(range(-fitsamplesize,fitsamplesize),contactpointleft[range(framenr-fitsamplesize,framenr+fitsamplesize)],1)
leftspeed[framenr]=leftposfit[0]
rightspeed[framenr]=rightposfit[0]
for fillinrest in range(0,fitsamplesize):
leftspeed[fillinrest]=leftspeed[fitsamplesize]
rightspeed[fillinrest]=rightspeed[fitsamplesize]
for fillinrest in range(nframes-fitsamplesize-1,nframes-1):
leftspeed[fillinrest]=leftspeed[nframes-fitsamplesize-1]
rightspeed[fillinrest]=rightspeed[nframes-fitsamplesize-1]
plt.close() #close the plot after we're done
elif nframes>1:
for framenr in range(0,nframes-2):
leftspeed[framenr]=contactpointleft[framenr+1]-contactpointleft[framenr]
rightspeed[framenr]=contactpointright[framenr+1]-contactpointright[framenr]
rightspeed[framenr-1]=rightspeed[framenr-2]
leftspeed[framenr-1]=leftspeed[framenr-2]
else:
leftspeed=0
rightspeed=0
return thetal, thetar, leftspeed, rightspeed, contactpointleft, contactpointright, dropvolume;
# Generate an icosphere
# -----------------------------------------------------------------------------
print('Generating icosphere')
icosphere = generate_icosphere(order)
# -----------------------------------------------------------------------------
# Load and process the cubemap image
# -----------------------------------------------------------------------------
print('Loading the cube map data')
# Load the multi-page TIFF image
# Channel 0: RGB
# Channel 1: Depth
# Channel 2: Sematic labels
# Channel 3: Instance labels
tiff = io.MultiImage('examples/inputs/cubemap.tiff')
# Convert RGB image to a torch tensor with dimensions (1, 3, H, W)
cube_rgb = torch.from_numpy(tiff[0]).permute(2, 0, 1).float().unsqueeze(0)
if cuda:
cube_rgb = cube_rgb.cuda()
# Convert depth image to torch tensor with dimensions (1, 1, H, W)
cube_inv_depth = torch.from_numpy(tiff[1].astype(
np.int32)).float().unsqueeze(0).unsqueeze(0)
if cuda:
cube_inv_depth = cube_inv_depth.cuda()
# Convert inverse depth to regular depth
cube_inv_depth[cube_inv_depth == 0] = -1
cube_depth = 1 / cube_inv_depth
