def plot_cup(img, bbox):
img = FC.resize_image(img)
lab_img = FC.apply_color_model(img)
labels1, labels2 = FC.segment_image(img, lab_img)
cup_labels = FC.select_cup_segment(labels2, bbox)
# out3 = skic.label2rgb(cup_labels, img, kind='avg')
out3 = cup_labels
neg_segments = FC.select_negative_segments(cup_labels, labels2)
out4 = skic.label2rgb(neg_segments, img, kind='avg')
plt.figure()
plt.imshow(img)
plt.title("Image")
plt.figure()
plt.imshow(skic.label2rgb(labels1, img, kind='avg'))
plt.title("Plain SLIC result")
plt.figure()
plt.imshow(skic.label2rgb(labels2, img, kind='avg'))
plt.title("Merged segments")
plt.figure()
plt.imshow(out3)
if np.max(out3.ravel()) == 1:
plt.title("cup only")
else:
plt.title("rejected cup")
plt.figure()
plt.imshow(out4)
plt.title("Negative segments")
def logo_iterate(labels, image, fns=d + 'logo-%03i.png'):
height, width = labels.shape
background = (labels == 0)
foreground = ~background
counter = it.count()
# part one: just foreground/background
colorcombos = it.permutations(colors, 2)
lab2 = np.zeros(labels.shape, np.uint8)
lab2[foreground] = 1
for cs in colorcombos:
img = color.label2rgb(lab2, image, colors=cs)
io.imsave(fns % next(counter), img)
# part two: background split
splits = np.arange(500, 1600, 100).astype(int)
colorcombos = it.permutations(colors, 3)
for s, cs in it.product(splits, colorcombos):
im, lab = _split_img_horizontal(image, lab2, background, s)
img = color.label2rgb(lab, im, colors=cs)
io.imsave(fns % next(counter), img)
# part three: foreground split
colorcombos = it.permutations(colors, 3)
for cs in colorcombos:
img = color.label2rgb(labels, image, colors=cs)
io.imsave(fns % next(counter), img)
# part four: both split
colorcombos = it.permutations(colors, 4)
for s, cs in it.product(splits, colorcombos):
im, lab = _split_img_horizontal(image, labels, background, s)
img = color.label2rgb(lab, im, colors=cs)
io.imsave(fns % next(counter), img)
def normcut_segmentations(img):
#labels1 = segmentation.slic(img, compactness=3, n_segments=50)
labels1 = segmentation.slic(img,compactness=3,n_segments=20)
out1 = color.label2rgb(labels1, img)#, kind='avg')
#return labels1
g = graph.rag_mean_color(img, labels1, mode='similarity')
labels2 = graph.cut_normalized(labels1, g)
out2 = color.label2rgb(labels2, img,image_alpha=0.2)#, kind='avg')
return (labels1,labels2)
def number_nucleus(image):
elevation_map = sobel(image)
markers = np.zeros_like(image)
markers[image < 250] = 1
markers[image > 2000] = 2
segmentation = watershed(elevation_map, markers)
label_img = label(segmentation)
prop = regionprops(label_img)
width, height = plt.rcParams['figure.figsize']
plt.rcParams['image.cmap'] = 'gray'
image_label_overlay = label2rgb(label_img, image=image)
fig, (ax1, ax2) = plt.subplots(ncols=2, nrows=1, figsize=(15, 8))
ax1.imshow(image_label_overlay)
ax2.imshow(image, cmap=plt.cm.gray, interpolation='nearest')
# create list of region with are < 1000
image_labeled = [region for region in prop if region.area > 5000]
return len(image_labeled)
def show_all(fname,images,titles,numsegs=1):
num_images = len(images)
num_titles = len(titles)
titles += ['']*(num_images-num_titles)
fig, axes = plt.subplots(ncols=num_images, figsize=(9, 2.5))
im = images[0]
for i in range(numsegs):
axes[i].imshow(images[i])
axes[i].set_title(titles[i])
print titles[i]
for i in range(numsegs,num_images) :
j=i #numsegs+i
segimg = label2rgb(images[j], image=im, image_alpha=0.5)
axes[j].imshow(segimg, interpolation='nearest')
axes[j].set_title(titles[j])
print titles[j]
for ax in axes:
ax.axis('off')
fig.subplots_adjust(hspace=0.01, wspace=0.01, top=1, bottom=0, left=0, right=1)
plt.show()
#plt.savefig(fname+"_seg.jpg")
print fname+"_seg.jpg"
return True
def build_region(self):
start_time = time.time();
labels = segmentation.slic(self.q_frame,self.num_superpixels, self.compactness,convert2lab=True,multichannel=True)
_num_superpixels = np.max(labels) + 1;
self.s_frame = color.label2rgb(labels,self.q_frame, kind='avg')
self.freq = np.array([np.sum(labels==label) for label in range(_num_superpixels)])
self.mean = np.array([region['centroid'] for region in regionprops(labels+1)],dtype=np.int16);
self.color_data = np.array([np.sum(self.q_frame[np.where(labels==label)],0) for label in range(_num_superpixels)])
_inv_freq = 1/(self.freq+0.0000001); self.color_data = self.color_data*_inv_freq[:,None]
gray_frame = cv2.cvtColor(self.q_frame,cv2.COLOR_RGB2GRAY)
def texture_prop(label,patch_size = 5):
_mean_min = self.mean[label]-patch_size;
_mean_max = self.mean[label]+patch_size;
glcm = greycomatrix(gray_frame[_mean_min[0]:_mean_max[0],_mean_min[1]:_mean_max[1]],
[3], [0], 256, symmetric=True, normed=True)
_dis = greycoprops(glcm, 'dissimilarity')[0, 0];
_cor = greycoprops(glcm, 'correlation')[0, 0];
return (_dis,_cor);
self.texture_data = np.array([texture_prop(label) for label in range(_num_superpixels)])
self.data = np.hstack((self.color_data,self.texture_data))
cv2.imwrite('outs.png',self.s_frame);
print "Build region (preprocess) : ",time.time()-start_time
return (labels,_num_superpixels);
def roofRegion(edge):
"""Estimate region based on edges of roofRegion
"""
# apply threshold
thresh = threshold_otsu(image)
bw = closing(image > thresh, square(3))
# remove artifacts connected to image border
cleared = bw.copy()
clear_border(cleared)
# label image regions
label_image = label(cleared)
borders = np.logical_xor(bw, cleared)
label_image[borders] = -1
image_label_overlay = label2rgb(label_image, image=image)
fig, ax = plt.subplots(ncols=1, nrows=1, figsize=(6, 6))
ax.imshow(image_label_overlay)
for region in regionprops(label_image):
# skip small images
if region.area < 100:
continue
# draw rectangle around segmented coins
minr, minc, maxr, maxc = region.bbox
rect = mpatches.Rectangle((minc, minr), maxc - minc, maxr - minr,
fill=False, edgecolor='red', linewidth=2)
ax.add_patch(rect)
plt.show()
def __build_region__(self,q_frame): start_time = time.time(); regions = segmentation.slic(q_frame,self.props.num_superpixels, self.props.compactness, convert2lab=self.props.useLAB,multichannel=True) num_regions = len(np.unique(regions)); s_frame = color.label2rgb(regions,q_frame, kind='avg') mean = np.array([region['centroid'] for region in regionprops(regions+1)]) freq = np.array([np.sum(regions==region) for region in range(num_regions)]) region_props = (mean,freq); if self.props.useColor: color_data = self.__extract_color__(q_frame,regions,region_props); if self.props.useTexture: texture_data = self.__extract_texture__(q_frame,regions,region_props); if self.props.useTexture and self.props.useColor: data = np.hstack((color_data,texture_data)) elif self.props.useTexture: data = texture_data else : data = color_data if self.props.doProfile: cv2.imwrite(self.PROFILE_PATH+self.method+'_s.png',s_frame); print "Build region (preprocess) : ",time.time()-start_time return (num_regions,regions,region_props,data);
def get_cells(image):
'''
Get cellls from the polygon.
'''
new_image=np.ones([3,image.shape[0],image.shape[1]],dtype=float)
# apply threshold
thresh = threshold_otsu(image)
bw=image
# remove artifacts connected to image border
cleared = bw.copy()
clear_border(cleared)
# label image regions
label_image = label(cleared)
#skimage.measure.label
#find_contours
borders = np.logical_xor(bw, cleared)
label_image[borders] = -1
image_label_overlay = label2rgb(label_image, image=image)
#extract the regions and get a polygon per region
polygons=[]
for i,region in enumerate(regionprops(label_image)):
# skip small images
if region.area < 100:
continue
#polygons.append(matplotlib.path.Path(region.coords))
print (region.coords.shape)
a=np.zeros(region.coords.shape)
a[:,0]=region.coords[:,1]
a[:,1]=region.coords[:,0]
polygons.append(a)
return polygons
def _apply(self, img_msg, label_msg):
bridge = cv_bridge.CvBridge()
img = bridge.imgmsg_to_cv2(img_msg)
label_img = bridge.imgmsg_to_cv2(label_msg)
# publish only valid label region
applied = img.copy()
applied[label_img == 0] = 0
applied_msg = bridge.cv2_to_imgmsg(applied, encoding=img_msg.encoding)
applied_msg.header = img_msg.header
self.pub_img.publish(applied_msg)
# publish visualized label
if img_msg.encoding in {'16UC1', '32SC1'}:
# do dynamic scaling to make it look nicely
min_value, max_value = img.min(), img.max()
img = (img - min_value) / (max_value - min_value) * 255
img = gray2rgb(img)
label_viz_img = label2rgb(label_img, img, bg_label=0)
label_viz_img = mark_boundaries(label_viz_img, label_img, (1, 0, 0))
label_viz_img = (label_viz_img * 255).astype(np.uint8)
label_viz_msg = bridge.cv2_to_imgmsg(label_viz_img, encoding='rgb8')
label_viz_msg.header = img_msg.header
self.pub_label_viz.publish(label_viz_msg)
# publish mask
if self._publish_mask:
bg_mask = (label_img == 0)
fg_mask = ~bg_mask
bg_mask = (bg_mask * 255).astype(np.uint8)
fg_mask = (fg_mask * 255).astype(np.uint8)
fg_mask_msg = bridge.cv2_to_imgmsg(fg_mask, encoding='mono8')
fg_mask_msg.header = img_msg.header
bg_mask_msg = bridge.cv2_to_imgmsg(bg_mask, encoding='mono8')
bg_mask_msg.header = img_msg.header
self.pub_fg_mask.publish(fg_mask_msg)
self.pub_bg_mask.publish(bg_mask_msg)
def detectOpticDisc(image):
labels = segmentation.slic(image, n_segments = 70)
out = color.label2rgb(labels, image, kind='avg')
gray = cv2.cvtColor(out, cv2.COLOR_RGB2GRAY)
minimum = np.max(gray)
image[gray==minimum] = 255
return image
def plot_preprocessed_image(self):
"""
plots pre-processed image. The plotted image is the same as obtained at the end
of the get_text_candidates method.
"""
image = restoration.denoise_tv_chambolle(self.image, weight=0.1)
thresh = threshold_otsu(image)
bw = closing(image > thresh, square(2))
cleared = bw.copy()
label_image = measure.label(cleared)
borders = np.logical_xor(bw, cleared)
label_image[borders] = -1
image_label_overlay = label2rgb(label_image, image=image)
fig, ax = plt.subplots(ncols=1, nrows=1, figsize=(12, 12))
ax.imshow(image_label_overlay)
for region in regionprops(label_image):
if region.area < 10:
continue
minr, minc, maxr, maxc = region.bbox
rect = mpatches.Rectangle((minc, minr), maxc - minc, maxr - minr,
fill=False, edgecolor='red', linewidth=2)
ax.add_patch(rect)
plt.show()
def SuperPixel(self, Image):
segments = slic(Image, n_segments=20, sigma=5)
# show the output of SLIC
segments = segments + 1
# So that no labelled region is 0 and ignored by regionprops
label_rgb = color.label2rgb(segments, Image, kind='avg')
return label_rgb
def detectOpticDisc(image):
kernel = octagon(10, 10)
thresh = threshold_otsu(image[:,:,1])
binary = image > thresh
print binary.dtype
luminance = convertToHLS(image)[:,:,2]
t = threshold_otsu(luminance)
t = erosion(luminance, kernel)
labels = segmentation.slic(image[:,:,1], n_segments = 3)
out = color.label2rgb(labels, image[:,:,1], kind='avg')
skio.imshow(out)
x, y = computeCentroid(t)
print x, y
rows, cols, _ = image.shape
p1 = closing(image[:,:,1],kernel)
p2 = opening(p1, kernel)
p3 = reconstruction(p2, p1, 'dilation')
p3 = p3.astype(np.uint8)
#g = dilation(p3, kernel)-erosion(p3, kernel)
#g = rank.gradient(p3, disk(5))
g = cv2.morphologyEx(p3, cv2.MORPH_GRADIENT, kernel)
#markers = rank.gradient(p3, disk(5)) < 10
markers = drawCircle(rows, cols, x, y, 85)
#markers = ndimage.label(markers)[0]
#skio.imshow(markers)
g = g.astype(np.uint8)
#g = cv2.cvtColor(g, cv2.COLOR_GRAY2RGB)
w = watershed(g, markers)
print np.max(w), np.min(w)
w = w.astype(np.uint8)
#skio.imshow(w)
return w
def view_dataset(self):
for datum in self.val:
rgb, label = self.load_datum(datum, train=False)
label_viz = label2rgb(label, rgb, bg_label=-1)
label_viz[label == 0] = 0
plt.imshow(label_viz)
plt.show()
def get_cells(image):
'''
Get cellls from the polygon.
'''
# apply threshold
thresh = threshold_otsu(image)
binary = image > thresh
bw=binary
plt.imshow(bw)
# Remove connected to image border
cleared = bw.copy()
clear_border(cleared)
# label image regions
label_image = skimage.measure.label(cleared)
#find_contours
borders = np.logical_xor(bw, cleared)
label_image[borders] = -1
image_label_overlay = label2rgb(label_image, image=image)
#extract the regions and get a polygon per region
polygons=[]
for i,region in enumerate(regionprops(label_image)):
# skip small images
if region.area < 100:
continue
a=np.zeros([len(region.coords),2])
#a=np.zeros(
plt.imshow(bw)
for i in range(len(region.coords)):
a[i,:]=[region.coords[i][0],region.coords[i][1]]
polygons.append(a)
return polygons
def getRegions():
"""Geocode address and retreive image centered
around lat/long"""
address = request.args.get('address')
results = Geocoder.geocode(address)
lat, lng = results[0].coordinates
zip_code = results[0].postal_code
map_url = 'https://maps.googleapis.com/maps/api/staticmap?center={0},{1}&size=640x640&zoom=19&sensor=false&maptype=roadmap&&style=visibility:simplified|gamma:0.1'
request_url = map_url.format(lat, lng)
req = urllib.urlopen(request_url)
img = io.imread(req.geturl(),flatten=True)
labels, numobjects = ndimage.label(img)
image = filter.canny(img, sigma=3)
thresh = threshold_otsu(image)
bw = closing(image > thresh, square(3))
# remove artifacts connected to image border
cleared = bw.copy()
clear_border(cleared)
# label image regions
label_image = label(cleared)
borders = np.logical_xor(bw, cleared)
label_image[borders] = -1
image_label_overlay = label2rgb(label_image, image=image)
fig, ax = plt.subplots(ncols=1, nrows=1, figsize=(6, 6))
ax.imshow(image_label_overlay)
def _callback(self, img_msg, mask_msg):
bridge = cv_bridge.CvBridge()
bgr_img = bridge.imgmsg_to_cv2(img_msg, desired_encoding='bgr8')
mask_img = bridge.imgmsg_to_cv2(mask_msg, desired_encoding='mono8')
if mask_img.size < 1:
logwarn_throttle(10, 'Too small sized image')
return
logwarn_throttle(10, '[FCNMaskForLabelNames] >> Start Processing <<')
if mask_img.ndim == 3 and mask_img.shape[2] == 1:
mask_img = mask_img.reshape(mask_img.shape[:2])
if mask_img.shape != bgr_img.shape[:2]:
jsk_logwarn('Size of mask and color image is different.'
'Resizing.. mask {0} to {1}'
.format(mask_img.shape, bgr_img.shape[:2]))
mask_img = resize(mask_img, bgr_img.shape[:2],
preserve_range=True).astype(np.uint8)
blob = bgr_img - self.mean_bgr
blob = blob.transpose((2, 0, 1))
x_data = np.array([blob], dtype=np.float32)
if self.gpu != -1:
x_data = cuda.to_gpu(x_data, device=self.gpu)
x = Variable(x_data, volatile=True)
self.model(x)
pred_datum = cuda.to_cpu(self.model.score.data[0])
candidate_labels = [self.target_names.index(name)
for name in self.tote_contents]
label_pred_in_candidates = pred_datum[candidate_labels].argmax(axis=0)
label_pred = np.zeros_like(label_pred_in_candidates)
for idx, label_val in enumerate(candidate_labels):
label_pred[label_pred_in_candidates == idx] = label_val
label_pred[mask_img == 0] = 0 # set bg_label
label_viz = label2rgb(label_pred, bgr_img, bg_label=0)
label_viz = (label_viz * 255).astype(np.uint8)
debug_msg = bridge.cv2_to_imgmsg(label_viz, encoding='rgb8')
debug_msg.header = img_msg.header
self.pub_debug.publish(debug_msg)
output_mask = np.ones(mask_img.shape, dtype=np.uint8)
output_mask *= 255
for label_val, label_name in enumerate(self.target_names):
if label_name in self.label_names:
assert label_name == 'kleenex_paper_towels'
assert label_val == 21
label_mask = ((label_pred == label_val) * 255).astype(np.uint8)
contours, hierachy = cv2.findContours(
label_mask, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
cv2.drawContours(output_mask, contours, -1, 255, -1)
# output_mask[label_pred == label_val] = False
# output_mask = output_mask.astype(np.uint8)
# output_mask[output_mask == 1] = 255
output_mask[mask_img == 0] = 0
output_mask_msg = bridge.cv2_to_imgmsg(output_mask, encoding='mono8')
output_mask_msg.header = img_msg.header
self.pub.publish(output_mask_msg)
logwarn_throttle(10, '[FCNMaskForLabelNames] >> Finshed processing <<')
def validate(self):
"""Validate training with data."""
log_templ = ('{i_iter}: type={type}, loss={loss}, acc={acc}, '
'acc_cls={acc_cls}, iu={iu}, fwavacc={fwavacc}')
type = 'val'
self.model.train = False
N_data = len(self.dataset.val)
result = defaultdict(list)
desc = '{0}: validating'.format(self.i_iter)
for indice in tqdm.tqdm(xrange(N_data), ncols=80, desc=desc):
loss, acc, acc_cls, iu, fwavacc = self._iterate_once(
type=type, indices=[indice])
result['loss'].append(loss)
result['acc'].append(acc)
result['acc_cls'].append(acc_cls)
result['iu'].append(iu)
result['fwavacc'].append(fwavacc)
# visualize predicted label
blob = cuda.to_cpu(self.model.x.data)[0]
label_true = cuda.to_cpu(self.model.t.data)[0]
img = self.dataset.datum_to_img(blob)
label_true_viz = label2rgb(label_true, img, bg_label=0)
label_true_viz[label_true == 0] = 0
label_true_viz = (label_true_viz * 255).astype(np.uint8)
label = cuda.to_cpu(self.model.score.data)[0].argmax(axis=0)
label_viz = label2rgb(label, img, bg_label=0)
label_viz[label == 0] = 0
label_viz = (label_viz * 255).astype(np.uint8)
hline = np.zeros((5, img.shape[1], 3), dtype=np.uint8)
hline.fill(255)
imsave(
osp.join(self.log_dir, 'visualize_{0}.jpg'.format(self.i_iter)),
np.vstack([img, hline, label_true_viz, hline, label_viz, hline]))
log = dict(
i_iter=self.i_iter,
type=type,
loss=np.array(result['loss']).mean(),
acc=np.array(result['acc']).mean(),
acc_cls=np.array(result['acc_cls']).mean(),
iu=np.array(result['iu']).mean(),
fwavacc=np.array(result['fwavacc']).mean(),
)
print(log_templ.format(**log))
self.logfile.write(
'{i_iter},{type},{loss},{acc},{acc_cls},{iu},{fwavacc}\n'
.format(**log))
def forward(self, bottom, top):
# bottom[0]: images N*3*W*H
# bottom[1]: prediction N*1*W*H
n = bottom[0].data.shape[0]
for i in range(n):
labels = segmentation.slic( bottom[0].data[i].transpose((1,2,0)),
compactness=self.compactness, n_segments=self.n_segs)
top[0].data[i, ...] = color.label2rgb(labels, bottom[1].data[i].transpose((1,2,0)), kind='avg').transpose((2,0,1)) #.reshape(top[0].data[i].shape)
def fig_label_segments(fig, image, segments, label):
labels, _ = ndimage.label(segments)
image_label_overlay=label2rgb(labels, image=image)
image_label_overlay=mark_boundaries(image, segments)
fig.set_title(label)
fig.axis('off')
fig.imshow(image_label_overlay)
print ("%s number of segments: %d" % (label, len(np.unique(segments))))
def agglomerativeClusteringFeatures(image):
connectivity = grid_to_graph(*image[:,:,2].shape)
X = np.reshape(image[:,:,2], (-1,1))
ward = AgglomerativeClustering(n_clusters=150,
linkage = 'ward', connectivity = connectivity).fit(X)
labels = np.reshape(ward.labels_, image[:,:,2].shape)
averageIntensity = color.label2rgb(labels, image[:,:,2], kind = 'avg')
#areas = getAreas(labels)
return averageIntensity
def build_region(self):
start_time = time.time()
labels = segmentation.slic(
self.q_cur_frame, self.num_superpixels, self.compactness, convert2lab=True, multichannel=True
)
self.s_frame = color.label2rgb(labels, self.q_cur_frame, kind="avg")
cv2.imwrite("outs.png", self.s_frame)
print "Slic time : ", time.time() - start_time
return labels
def segment_image(image, n_segments=400, compactness=30, sigma=5, verbose=True):
if verbose:
print("segmenting image")
print("n_segments=%d, compactness=%d, sigma=%d" % (n_segments, compactness, sigma))
image = image.astype("float64")
labels = slic(image, compactness=compactness, n_segments=n_segments, sigma=sigma)
segmented = color.label2rgb(labels, image, kind="avg")
return segmented.astype("uint8")
def spectral_cluster(filename, compactness_val=30, n=6):
img = misc.imread(filename)
labels1 = segmentation.slic(img, compactness=compactness_val, n_segments=n)
out1 = color.label2rgb(labels1, img, kind='overlay', colors=['red','green','blue','cyan','magenta','yellow'])
fig, ax = plt.subplots()
ax.imshow(out1, interpolation='nearest')
ax.set_title("Compactness: {} | Segments: {}".format(compactness_val, n))
plt.show()
def createImage(thres):
# do simple filter based on color value
filtered_image = np.zeros_like(image) # set up all-zero image
filtered_image[image > thres] = 1 # filtered values set to 1
# label features and convert to rgb image
labeled_particles, num_features = ndi.label(filtered_image)
image_label_overlay = label2rgb(labeled_particles, image=image, bg_label=0)
return image_label_overlay
def getArea(address):
"""Geocode address and retreive image centered
around lat/long"""
address = address
results = Geocoder.geocode(address)
lat, lng = results[0].coordinates
zip_code = results[0].postal_code
map_url = 'https://maps.googleapis.com/maps/api/staticmap?center={0},{1}&size=640x640&zoom=19&sensor=false&maptype=roadmap&&style=visibility:simplified|gamma:0.1'
request_url = map_url.format(lat, lng)
req = urllib.urlopen(request_url)
img = io.imread(req.geturl(),flatten=True)
labels, numobjects = ndimage.label(img)
image = filter.canny(img, sigma=3)
thresh = threshold_otsu(image)
bw = closing(image > thresh, square(3))
# remove artifacts connected to image border
cleared = bw.copy()
clear_border(cleared)
# label image regions
label_image = label(cleared)
borders = np.logical_xor(bw, cleared)
label_image[borders] = -1
image_label_overlay = label2rgb(label_image, image=image)
fig, ax = plt.subplots(ncols=1, nrows=1, figsize=(6, 6))
ax.imshow(image_label_overlay)
dist = []
rp = regionprops(label_image)
rp = [x for x in rp if 100 < x.area <= 900]
for region in rp:
# skip small images
#if region.area < 100:
# continue
dist.append(sqrt( ( 320-region.centroid[0] )**2 + ( 320-region.centroid[1] )**2 ))
# draw rectangle around segmented coins
#minr, minc, maxr, maxc = region.bbox
#rect = mpatches.Rectangle((minc, minr), maxc - minc, maxr - minr,
# fill=False, edgecolor='red', linewidth=2)
#ax.add_patch(rect)
roof_index = dist.index(min(dist))
minr, minc, maxr, maxc = rp[roof_index].bbox
rect = mpatches.Rectangle((minc, minr), maxc - minc, maxr - minr,
fill=False, edgecolor='red', linewidth=2)
ax.add_patch(rect)
img = StringIO()
fig.savefig(img)
img.seek(0)
session['roof_area'] = rp[roof_index].area
roof_area = (rp[roof_index].area)*12
return(roof_area)
def overlay_cells_on_image(image, boundries, cells, offset=(0,0)):
mask = numpy.zeros(image.shape[:2])
boundries -= numpy.asarray(tuple(offset) * 2)
label = 1
for cell in filter_cells_by_boundry(cells, boundries):
math.modify_with_bounding_box(cell.bbox, mask, cell.mask.astype(numpy.uint16)*label)
label += 1
return label2rgb(mask, image=image, bg_label=0)
def parse_image(self):
self.m_print("Parsing panel image",0)
thresh = threshold_otsu(self.i_file)
bw = closing(self.i_file > thresh, square(3))
# remove artifacts connected to image border
cleared = bw.copy()
clear_border(cleared)
# label image regions
label_image = label(cleared)
borders = np.logical_xor(bw, cleared)
label_image[borders] = -1
return label2rgb(label_image, image=self.i_file),label_image
def build_region(self):
start_time = time.time();
labels = segmentation.slic(self.q_cur_frame,self.num_superpixels, self.compactness,convert2lab=True,multichannel=True)
_num_superpixels = np.max(labels) + 1;
self.s_frame = color.label2rgb(labels,self.q_cur_frame, kind='avg')
self.mean = np.array([region['centroid'] for region in regionprops(labels+1)])
self.color_data = np.array([np.sum(self.q_cur_frame[np.where(labels==label)],0) for label in range(_num_superpixels)])
self.freq = np.array([np.sum(labels==label) for label in range(_num_superpixels)])
_inv_freq = 1/(self.freq+0.0000001); self.color_data = self.color_data*_inv_freq[:,None]
cv2.imwrite('outs.png',self.s_frame);
print "Build region (preprocess) : ",time.time()-start_time
return (labels,_num_superpixels);
def image_pre_process(image):
# Apply otsu filter to find the thresholds
print('starting with multi otsu filter')
thresholds = threshold_multiotsu(image, classes=THO_N + 1)
print('thresholds are', thresholds)
regions = np.digitize(image, bins=thresholds)
otsu_regions_img = img_as_ubyte(regions)
# Binary threshold methods
# https://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_imgproc/py_thresholding/py_thresholding.html
ret, img_binary = cv2.threshold(image, thresholds[0], 255,
cv2.THRESH_BINARY)
# Dilation and erosion
# https://www.youtube.com/watch?v=WQK_oOWW5Zo
kernel = np.ones((3, 3), np.uint8)
erosion = cv2.erode(img_binary, kernel, iterations=1)
dilation = cv2.dilate(img_binary, kernel, iterations=1)
opening = cv2.morphologyEx(img_binary, cv2.MORPH_OPEN, kernel)
# Label image
# https://www.youtube.com/watch?v=u3nG5_EjfM0&list=PLZsOBAyNTZwbIjGnolFydAN33gyyGP7lT&index=119
#label_image = measure.label(otsu_regions_img, connectivity=img_binary.ndim)
label_image = measure.label(opening, connectivity=img_binary.ndim)
image_label_overlay = label2rgb(label_image, image=otsu_regions_img)
# Get different regions from the labeled image
# https://scikit-image.org/docs/dev/auto_examples/segmentation/plot_label.html
i = 0
regions = []
bbx = []
for region in measure.regionprops(label_image):
i += 1
print('i is', i)
print('region is', region.image)
regions.append(region.image)
bbx.append(region.bbox)
figure = plt.figure()
plt.imshow(region.image)
plt.title('region ' + np.str(i))
plt.savefig(
'G:\\My Drive\\Project\\IntraOral Scanner Registration\\Results\\Segmentation test\\region '
+ np.str(i))
plt.close()
print('shape of regions is', np.shape(regions))
for bbx_ind in bbx:
#figure = plt.figure()
#plt.imshow(regions[-3])
#plt.title('label image check')
print('region bounding box is', bbx_ind)
# Edge detection
# https://www.youtube.com/watch?v=Oy4duAOGdWQ&list=PLZsOBAyNTZwbIjGnolFydAN33gyyGP7lT&index=105
robert_image = roberts(img_binary)
sobel_image = sobel(img_binary)
scharr_image = scharr(img_binary)
prewitt_image = prewitt(img_binary)
farid_image = farid(img_binary)
print('detected edge is', sobel_image)
return otsu_regions_img, image_label_overlay, sobel_image
def run(self, ips, snap, img, para=None):
lab = segmentation.felzenszwalb(snap, para['scale'], para['sigma'],
para['min_size'])
return color.label2rgb(lab, snap, kind='avg')
def run(self, ips, snap, img, para=None):
lab = segmentation.quickshift(snap, para['ratio'], para['kernel_size'],
para['max_dist'], para['sigma'])
return color.label2rgb(lab, snap, kind='avg')
def labelsConnection(data):
labels = measure.label(data,
connectivity=2) #connectivity表示连接的模式，1代表4连通，2代表8连通
dst = color.label2rgb(labels) #根据不同的标记显示不同的颜色
#print('regions number:',labels.max()+1) #显示连通区域块数(从0开始标记)
return dst
So you will reduce this image from 265×191=50,615 pixels down to 400 regions. Young woman Already preloaded as face_image. The show_image() function has been preloaded for you as well. Instructions 100 XP Import the slic() function from the segmentation module. Import the label2rgb() function from the color module. Obtain the segmentation with 400 regions using slic(). Put segments on top of original image to compare with label2rgb(). ''' SOLUTION # Import the slic function from segmentation module from skimage.segmentation import slic # Import the label2rgb function from color module from skimage.color import label2rgb # Obtain the segmentation with 400 regions segments = slic(face_image, n_segments=400) # Put segments on top of original image to compare segmented_image = label2rgb(segments, face_image, kind='avg') # Show the segmented image show_image(segmented_image, "Segmented image, 400 superpixels")
# -*- coding: utf-8 -*-
"""
Created on Mon Sep 2 22:18:57 2019
@author: dtket
"""
import imageio as iio
from skimage import filters
from skimage.color import rgb2gray # only needed for incorrectly saved images
from skimage.measure import regionprops
import matplotlib.pyplot as plt
from skimage.color import label2rgb
image = rgb2gray(iio.imread('cube.png'))
threshold_value = filters.threshold_otsu(image)
labeled_foreground = (image > threshold_value).astype(int)
properties = regionprops(labeled_foreground, image)
center_of_mass = properties[0].centroid
weighted_center_of_mass = properties[0].weighted_centroid
colorized = label2rgb(labeled_foreground, image, colors=['black'], alpha=0.1)
fig, ax = plt.subplots()
ax.imshow(colorized)
# Note the inverted coordinates because plt uses (x, y) while NumPy uses (row, column)
ax.scatter(center_of_mass[1], center_of_mass[0], s=160, c='C0', marker='+')
plt.show()
from skimage import data, segmentation, color from skimage.future import graph from matplotlib import pyplot as plt img = data.coffee() labels1 = segmentation.slic(img, compactness=30, n_segments=400) out1 = color.label2rgb(labels1, img, kind='avg') g = graph.rag_mean_color(img, labels1, mode='similarity') labels2 = graph.cut_normalized(labels1, g) out2 = color.label2rgb(labels2, img, kind='avg') plt.imshow(out1 ) plt.show()
icount = len(image_set)
start_time = datetime.datetime.now()
#开始去掉一些噪音
for ii, image_name in enumerate(image_set):
# image_name='003.png'
image = cv2.imread(os.path.join(image_predict_dir, image_name))
image2 = image[:, :, 0] # 只取一个分量即可
image2[image2 > 0] = 1 # 原始数据的掩模，非0的有很多种数，而模型的输入要求是二值的
# 采用skimage中的measure，寻找每一个连通区域
labeled_img, num = measure.label(image2, background=0, return_num=True)
dst = color.label2rgb(labeled_img)
classes = np.unique(labeled_img)
classes = classes[1:] # 不要背景0这一类
for c in classes:
inds = np.where(labeled_img == c)
inds = np.array(inds).T
# 忽略太少的点组成的连通区域
if len(inds[:, 1]) < 1500:
image2[labeled_img == c] = 0
else:
trans_pca = PCA(n_components=2).fit(inds) # 对该连通区域所有点的坐标集合，进行PCA变换
pcas = trans_pca.components_ # PCA的两个主成分，即为点的坐标集合的主要方向和次要方向
# 最主要的方向，计算方向角度theta，准备旋转theta角变成水平方向
pca1 = pcas[0]
ax = axes.ravel()
ax[0].imshow(camera, cmap=cm.gray)
ax[0].set_title('Input image')
ax[1].imshow(edges, cmap=cm.gray)
ax[1].set_title('Canny edges')
ax[2].imshow(edges * 0)
for line in lines:
p0, p1 = line
ax[2].plot((p0[0], p1[0]), (p0[1], p1[1]))
ax[2].set_xlim((0, camera2.shape[1]))
ax[2].set_ylim((camera2.shape[0], 0))
ax[2].set_title('Probabilistic Hough')
'''
camera = label2rgb(camera2, image=camera1)
fig, ax = plt.subplots(figsize=(10, 6))
ax.imshow(camera, cmap='gray', interpolation='nearest')
#ax.imshow(camera , cmap='gray', interpolation='nearest')
count = 0
regions = regionprops(camera2)
for region in regions:
# take regions with large enough areas
if region.area >= 650:
# draw rectangle around segmented coins
minr, minc, maxr, maxc = region.bbox
slice_hei = int((maxr - minr) * 0.1)
#print(minr, minc, maxr, maxc)
croped = camera1[minr - slice_hei:maxr + slice_hei, minc:maxc]
binary_crop = cam_clear[minr - slice_hei:maxr + slice_hei, minc:maxc]
#print("Here",camera.shape)
# Binary threshold methods
# https://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_imgproc/py_thresholding/py_thresholding.html
ret, img_binary = cv2.threshold(img_segment, thresholds[0], 255,
cv2.THRESH_BINARY)
# Dilation and erosion
# https://www.youtube.com/watch?v=WQK_oOWW5Zo
kernel = np.ones((3, 3), np.uint8)
erosion = cv2.erode(img_binary, kernel, iterations=1)
dilation = cv2.dilate(img_binary, kernel, iterations=1)
# Label image
# https://www.youtube.com/watch?v=u3nG5_EjfM0&list=PLZsOBAyNTZwbIjGnolFydAN33gyyGP7lT&index=119
label_image = measure.label(output, connectivity=img_binary.ndim)
image_label_overlay = label2rgb(label_image, image=output)
# Edge detection
# https://www.youtube.com/watch?v=Oy4duAOGdWQ&list=PLZsOBAyNTZwbIjGnolFydAN33gyyGP7lT&index=105
robert_image = roberts(img_binary)
sobel_image = sobel(img_binary)
scharr_image = scharr(img_binary)
prewitt_image = prewitt(img_binary)
farid_image = farid(img_binary)
#img_mutso = multiOtsu(3, img)
plt.figure()
plt.imshow(img_gray, cmap='gray')
plt.title("Grayscale original image")
plt.figure()
def augment_and_show(aug,
image,
mask=None,
bboxes=[],
categories=[],
category_id_to_name=[],
filename=None,
font_scale_orig=0.35,
font_scale_aug=0.35,
show_title=True,
**kwargs):
augmented = aug(image=image,
mask=mask,
bboxes=bboxes,
category_id=categories)
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
image_aug = cv2.cvtColor(augmented['image'], cv2.COLOR_BGR2RGB)
for bbox in bboxes:
visualize_bbox(image, bbox, **kwargs)
for bbox in augmented['bboxes']:
visualize_bbox(image_aug, bbox, **kwargs)
if show_title:
for bbox, cat_id in zip(bboxes, categories):
visualize_titles(image,
bbox,
category_id_to_name[cat_id],
font_scale=font_scale_orig,
**kwargs)
for bbox, cat_id in zip(augmented['bboxes'], augmented['category_id']):
visualize_titles(image_aug,
bbox,
category_id_to_name[cat_id],
font_scale=font_scale_aug,
**kwargs)
if mask is None:
f, ax = plt.subplots(1, 2, figsize=(16, 8))
ax[0].imshow(image)
ax[0].set_title('Original image')
ax[1].imshow(image_aug)
ax[1].set_title('Augmented image')
else:
f, ax = plt.subplots(2, 2, figsize=(16, 16))
if len(mask.shape) != 3:
mask = label2rgb(mask, bg_label=0)
mask_aug = label2rgb(augmented['mask'], bg_label=0)
else:
mask = cv2.cvtColor(mask, cv2.COLOR_BGR2RGB)
mask_aug = cv2.cvtColor(augmented['mask'], cv2.COLOR_BGR2RGB)
ax[0, 0].imshow(image)
ax[0, 0].set_title('Original image')
ax[0, 1].imshow(image_aug)
ax[0, 1].set_title('Augmented image')
ax[1, 0].imshow(mask, interpolation='nearest')
ax[1, 0].set_title('Original mask')
ax[1, 1].imshow(mask_aug, interpolation='nearest')
ax[1, 1].set_title('Augmented mask')
f.tight_layout()
if filename is not None:
f.savefig(filename)
return augmented['image'], augmented['mask'], augmented['bboxes']
def image_segmentation(in_file_name, out_file_name, show_image):
#example_ni1 = os.path.join(data_path, in_file_name)
n1_img = nib.load(in_file_name)
img_data = n1_img.get_data()
print(img_data.shape)
#save_example_ni1 = os.path.join(data_path, out_file_name)
slice = np.zeros((176, 176, 208))
segm = np.zeros((176, 176, 208))
for i in range(175):
slice[i] = img_data[:, :, i, 0]
slice[i] = exposure.rescale_intensity(slice[i], out_range=(0, 256))
img = color.gray2rgb(slice[i])
if (img.min() >= 0):
labels1 = segmentation.slic(img,
compactness=30,
n_segments=200,
multichannel=False)
out1 = color.label2rgb(labels1, img, kind='avg')
#g = graph.rag_mean_color(img, labels1, mode='similarity')
#labels2 = graph.cut_normalized(labels1, g)
#out2 = color.label2rgb(labels2, img, kind='avg')
segm[i] = color.rgb2gray(out1)
#segm[i] = out1
if (show_image):
show_slices([slice[100], slice[110], slice[120]])
plt.suptitle("slices")
for i in range(175):
img_data[:, :, i, 0] = segm[i]
if (show_image):
# display results
fig, (ax1, ax2, ax3) = plt.subplots(nrows=1,
ncols=3,
figsize=(8, 3),
sharex=True,
sharey=True)
ax1.imshow(img_data[:, :, 100, 0])
ax1.axis('off')
ax1.set_title('image 100', fontsize=20)
ax2.imshow(img_data[:, :, 110, 0])
ax2.axis('off')
ax2.set_title('image 110', fontsize=20)
ax3.imshow(img_data[:, :, 120, 0])
ax3.axis('off')
ax3.set_title('image 120', fontsize=20)
fig.subplots_adjust(wspace=0.02,
hspace=0.02,
top=0.9,
bottom=0.02,
left=0.02,
right=0.98)
plt.show()
save_img = nib.Nifti1Image(img_data, np.eye(4))
nib.save(save_img, save_example_ni1)
import numpy as np
from skimage import data, util, filters, color
from skimage.segmentation import watershed
import matplotlib.pyplot as plt
coins = data.coins()
edges = filters.sobel(coins)
grid = util.regular_grid(coins.shape, n_points=468)
seeds = np.zeros(coins.shape, dtype=int)
seeds[grid] = np.arange(seeds[grid].size).reshape(seeds[grid].shape) + 1
w0 = watershed(edges, seeds)
w1 = watershed(edges, seeds, compactness=0.01)
fig, (ax0, ax1) = plt.subplots(1, 2)
ax0.imshow(color.label2rgb(w0, coins, bg_label=-1))
ax0.set_title('Classical watershed')
ax1.imshow(color.label2rgb(w1, coins, bg_label=-1))
ax1.set_title('Compact watershed')
plt.show()
def extract(self, data):
path = data.basePath + "\\" + data.name
print(path)
img = cv2.imread(path, cv2.IMREAD_GRAYSCALE)
# print(img)
treshold = cv2.threshold(img, 127, 255, cv2.THRESH_BINARY)[1]
dots = treshold > treshold.mean()
dots_labels = measure.label(dots, background=1)
image_label_overlay = label2rgb(dots_labels, image=treshold)
max_area = 0
total_area = 0
count_connected_group = 0
average = 0.0
for region in regionprops(dots_labels):
if (region.area > 10):
total_area += region.area
count_connected_group += 1
if (region.area >= 250):
if (region.area > max_area):
max_area = region.area
average = (total_area / count_connected_group)
a4_constant = ((average / 84.0) * 250.0) + 100
b = morphology.remove_small_objects(dots_labels, a4_constant)
if os.path.isdir(data.basePath + '\\outputs'):
pass
else:
os.mkdir(data.basePath + '\\outputs')
plt.imsave(data.basePath + '\\outputs\\pre_version.png', b)
img = cv2.imread(data.basePath + '\\outputs\\pre_version.png',
cv2.IMREAD_GRAYSCALE)
img = cv2.threshold(img, 0, 255,
cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)[1]
cv2.imwrite(data.basePath + "\\outputs\\output.png", img)
print('Signature Extraction Success!')
image = cv2.imread(data.basePath + '\\outputs\\output.png')
result = image.copy()
image = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
lower = np.array([0, 0, 0])
upper = np.array([255, 255, 200])
mask = cv2.inRange(image, lower, upper)
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))
close = cv2.morphologyEx(mask, cv2.MORPH_CLOSE, kernel, iterations=1)
contours = cv2.findContours(close, cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)
if len(contours) == 2:
contours = contours[0]
else:
contours = contours[1]
boxes = []
for c in contours:
(x, y, w, h) = cv2.boundingRect(c)
boxes.append([x, y, x + w, y + h])
boxes = np.asarray(boxes)
left = np.min(boxes[:, 0])
top = np.min(boxes[:, 1])
right = np.max(boxes[:, 2])
bottom = np.max(boxes[:, 3])
result[close == 0] = (255, 255, 255)
ROI = result[top:bottom, left:right].copy()
path = data.basePath + '\\result_' + data.name
cv2.imwrite(path, ROI)
# print('Signature Capture Success')
return path
# Voila!
#
# This solution works reasonably well. Of course there is still a lot of room for improvement.
#
# For cells that get oversegmented we should implement a function that glues those oversegmented cells together rather than drop them.
#
# But that I will leave for you to create!
#
# If you liked the solution and would like to see how it works as a part of the end-to-end pipeline please go to:
#
# https://github.com/neptune-ml/data-science-bowl-2018
#
# To stay up to date with new features and fully open sourced solution (with predictions) read our data science bowl journal thread https://www.kaggle.com/c/data-science-bowl-2018/discussion/47590
#
# Cheers and good luck!
# # LB Test set predictions
# In[102]:
test_masks = joblib.load('../input/test-predictions/test_masks.pkl')
# In[103]:
from skimage.color import label2rgb
for mask in test_masks[:5]:
plt.imshow(label2rgb(mask))
plt.show()
#LOCAL SOURCE image_rgb = io.imread('examples/leds_off.jpg')
image = rgb2gray(image_rgb)
# apply threshold
bw = closing(image > detect_th, square(3))
# remove artifacts connected to image border
cleared = clear_border(bw)
# label image regions
label_image = label(cleared)
# to make the background transparent, pass the value of `bg_label`,
# and leave `bg_color` as `None` and `kind` as `overlay`
image_label_overlay = label2rgb(label_image, image=image, bg_label=0)
fig, ax = plt.subplots(figsize=(10, 6))
ax.imshow(image_label_overlay)
leds = {} #led objects
for region in regionprops(label_image):
# take regions with large enough areas
if region.area > 1000:
# draw rectangle around segmented item
minr, minc, maxr, maxc = region.bbox
cropped = image_rgb[minr:maxr, minc:maxc]
skimage.io.imsave("temp/" + str(region.label) + ".jpg", cropped)
def cConexas2(image):
label_image = label(image)
image_label_overlay = label2rgb(image, image=image)
fig, ax = plt.subplots(figsize=(10, 6))
imageComponentesConexas = np.zeros(image.shape)
ax.imshow(image_label_overlay)
i = 0
regions = []
for region in regionprops(label_image):
# take regions with large enough areas
regions.append(region)
# draw rectangle around segmented coins
minr, minc, maxr, maxc = region.bbox
if (i == 0):
rect = mpatches.Rectangle((minc, minr),
maxc - minc,
maxr - minr,
fill=False,
edgecolor='red',
linewidth=2)
elif (i == 1):
rect = mpatches.Rectangle((minc, minr),
maxc - minc,
maxr - minr,
fill=False,
edgecolor='green',
linewidth=2)
elif (i == 2):
rect = mpatches.Rectangle((minc, minr),
maxc - minc,
maxr - minr,
fill=False,
edgecolor='blue',
linewidth=2)
elif (i == 3):
rect = mpatches.Rectangle((minc, minr),
maxc - minc,
maxr - minr,
fill=False,
edgecolor='yellow',
linewidth=2)
else:
rect = mpatches.Rectangle((minc, minr),
maxc - minc,
maxr - minr,
fill=False,
edgecolor='purple',
linewidth=2)
i = i + 1
ax.add_patch(rect)
imageComponentesConexas = newImage(imageComponentesConexas,
region.coords)
ax.set_axis_off()
plt.tight_layout()
plt.imshow(imageComponentesConexas, cmap='gray')
plt.title("CC Finales")
return imageComponentesConexas, regions
def MERGEIAMGES(self, displaygui, CH1, RGB_Channels):
if displaygui.Ch1CheckBox.isChecked() == True:
ch1_rgb = np.stack((CH1, ) * 3, axis=-1)
else:
ch1_rgb = np.zeros(RGB_Channels.shape, dtype=np.uint8)
All_Channels = cv2.addWeighted(ch1_rgb, 1, RGB_Channels, 1, 0)
height, width, ch = np.shape(All_Channels)
totalBytes = All_Channels.nbytes
#print(self.AnalysisGui.NucleiChannel.currentIndex().dtype)
if displaygui.NuclMaskCheckBox.isChecked() == True:
self.input_image = self.IMAGE_TO_BE_MASKED(displaygui)
bound, filled_res = ImageAnalyzer.neuceli_segmenter(
self.input_image)
#cv2.imwrite('mask_saved.jpg',bound)
if displaygui.NucPreviewMethod.currentText() == "Boundary":
All_Channels[bound != 0] = [255, 0, 0]
if displaygui.NucPreviewMethod.currentText() == "Area":
labeled_array, num_features = label(filled_res)
rgblabel = label2rgb(labeled_array, alpha=0.1, bg_label=0)
rgblabel = cv2.normalize(rgblabel,
None,
0,
255,
cv2.NORM_MINMAX,
dtype=cv2.CV_8U)
image_input_stack = np.stack((self.input_image, ) * 3, axis=-1)
All_Channels = cv2.addWeighted(rgblabel, 0.2, ch1_rgb, 1, 0)
##############
if displaygui.SpotsCheckBox.isChecked() == True:
ch1_spots_log, ch2_spots_log, ch3_spots_log, ch4_spots_log = self.IMAGE_FOR_SPOT_DETECTION(
displaygui)
if displaygui.SpotPreviewMethod.currentText() == "Dots":
if ch1_spots_log != []:
All_Channels[ch1_spots_log != 0] = [255, 255, 255]
if ch2_spots_log != []:
All_Channels[ch2_spots_log != 0] = [255, 0, 0]
if ch3_spots_log != []:
All_Channels[ch3_spots_log != 0] = [0, 255, 0]
if ch4_spots_log != []:
All_Channels[ch4_spots_log != 0] = [0, 0, 255]
if displaygui.NucPreviewMethod.currentText() == "Cross":
pass
self.SHOWIMAGE(displaygui, All_Channels, width, height, totalBytes)
def _callback(self, img_msg, mask_msg):
bridge = cv_bridge.CvBridge()
bgr_img = bridge.imgmsg_to_cv2(img_msg, desired_encoding='bgr8')
mask_img = bridge.imgmsg_to_cv2(mask_msg, desired_encoding='mono8')
if mask_img.size < 1:
logwarn_throttle(10, 'Too small sized image')
return
logwarn_throttle(10, '[FCNMaskForLabelNames] >> Start Processing <<')
if mask_img.ndim == 3 and mask_img.shape[2] == 1:
mask_img = mask_img.reshape(mask_img.shape[:2])
if mask_img.shape != bgr_img.shape[:2]:
jsk_logwarn('Size of mask and color image is different.'
'Resizing.. mask {0} to {1}'.format(
mask_img.shape, bgr_img.shape[:2]))
mask_img = resize(mask_img, bgr_img.shape[:2],
preserve_range=True).astype(np.uint8)
blob = bgr_img - self.mean_bgr
blob = blob.transpose((2, 0, 1))
x_data = np.array([blob], dtype=np.float32)
if self.gpu != -1:
x_data = cuda.to_gpu(x_data, device=self.gpu)
x = Variable(x_data, volatile=True)
self.model(x)
pred_datum = cuda.to_cpu(self.model.score.data[0])
candidate_labels = [
self.target_names.index(name) for name in self.tote_contents
]
label_pred_in_candidates = pred_datum[candidate_labels].argmax(axis=0)
label_pred = np.zeros_like(label_pred_in_candidates)
for idx, label_val in enumerate(candidate_labels):
label_pred[label_pred_in_candidates == idx] = label_val
label_pred[mask_img == 0] = 0 # set bg_label
label_viz = label2rgb(label_pred, bgr_img, bg_label=0)
label_viz = (label_viz * 255).astype(np.uint8)
debug_msg = bridge.cv2_to_imgmsg(label_viz, encoding='rgb8')
debug_msg.header = img_msg.header
self.pub_debug.publish(debug_msg)
output_mask = np.ones(mask_img.shape, dtype=np.uint8)
output_mask *= 255
for label_val, label_name in enumerate(self.target_names):
if label_name in self.label_names:
assert label_name == 'kleenex_paper_towels'
assert label_val == 21
label_mask = ((label_pred == label_val) * 255).astype(np.uint8)
contours, hierachy = cv2.findContours(label_mask,
cv2.RETR_TREE,
cv2.CHAIN_APPROX_SIMPLE)
cv2.drawContours(output_mask, contours, -1, 255, -1)
# output_mask[label_pred == label_val] = False
# output_mask = output_mask.astype(np.uint8)
# output_mask[output_mask == 1] = 255
output_mask[mask_img == 0] = 0
output_mask_msg = bridge.cv2_to_imgmsg(output_mask, encoding='mono8')
output_mask_msg.header = img_msg.header
self.pub.publish(output_mask_msg)
logwarn_throttle(10, '[FCNMaskForLabelNames] >> Finshed processing <<')
def show(*obj, file_name=None, overlay=False, pred=False,
show_bbox=True, figsize=(10,10), cmap='binary_r', **kwargs):
"Show image, mask, and weight (optional)"
if len(obj)==3:
img,msk,weight = obj
elif len(obj)==2:
img,msk = obj
weight = None
elif len(obj)==1:
img = obj[0]
msk, weight = None, None
else:
raise ValueError(f'Function not defined for {len(obj)} arguments.')
# Image preprocessing
img = np.array(img)
# Swap axis to channels last
if img.shape[0]<20: img=np.moveaxis(img,0,-1)
# One channel images
if img.ndim == 3 and img.shape[-1] == 1:
img=img[...,0]
# Mask preprocessing
if msk is not None:
msk = np.array(msk)
# Remove background class from masks
if msk.shape[0]==2: msk=msk[1,...]
# Create bbox
pad = (np.array(img.shape[:2])-np.array(msk.shape))//2
bbox = Rectangle((pad[0]-1,pad[1]-1),img.shape[0]-2*pad[0]+1,img.shape[0]-2*pad[0]+1,
edgecolor='r',linewidth=1,facecolor='none')
# Padding mask and weights
msk = np.pad(msk, pad, 'constant', constant_values=(0))
if cmap is None:
cmap = 'binary_r' if msk.max()==1 else cmap
# Weights preprocessing
if weight is not None:
weight = np.array(weight)
weight = np.pad(weight, pad, 'constant', constant_values=(0))
ncol=1 if msk is None else 2
ncol=ncol if weight is None else ncol+1
fig, ax = plt.subplots(1,ncol,figsize=figsize)
img_ax = ax[0] if ncol>1 else ax
# Plot img
img_ax.imshow(img, cmap=cmap)
if file_name is not None:
img_ax.set_title('Image {}'.format(file_name))
else:
img_ax.set_title('Image')
img_ax.set_axis_off()
# Plot img and mask
if msk is not None:
if overlay:
label_image = label(msk)
img_l2o = label2rgb(label_image, image=img, bg_label=0, alpha=.8, image_alpha=1)
ax[1].set_title('Image + Mask (#ROIs: {})'.format(label_image.max()))
ax[1].imshow(img_l2o)
else:
ax[1].imshow(msk, cmap=cmap)
ax[1].set_title('Mask')
if show_bbox: ax[1].add_patch(copy(bbox))
ax[1].set_axis_off()
# Plot weights
if weight is not None:
max_w = weight.max()
vmax_w = max(1, max_w)
ax[2].imshow(weight, vmax=vmax_w, cmap=cmap)
if pred:
ax[2].set_title('Prediction')
else:
ax[2].set_title('Weights (max value: {:.{p}f})'.format(max_w, p=1))
if show_bbox: ax[2].add_patch(copy(bbox))
ax[2].set_axis_off()
#ax.set_axis_off()
plt.tight_layout()
plt.show()
axes.imshow(image, cmap="gray")
axes.set_title('Gray image')
print(image.shape)
global_thresh = threshold_otsu(image)
binary_global = image < global_thresh
fig, axes = plt.subplots(nrows=1, figsize=(40, 40))
axes.imshow(binary_global, cmap="gray")
axes.set_title('Binary image')
label_image, n = measure.label(binary_global,
neighbors=8,
background=0,
return_num=True)
image_label_overlay = label2rgb(label_image, image=binary_global)
fig, axes = plt.subplots(nrows=1, figsize=(40, 40))
axes.imshow(image_label_overlay)
axes.set_title('Conected components')
area_thresh = 10000 #@param {type:"slider", min:0, max:1000000, step:1}
seg_prop = 0.011 #@param {type:"slider", min:0.001, max:100.0, step:0.01}
selected_components = np.zeros(label_image.shape, dtype=int)
selected_components_slic = np.zeros(label_image.shape, dtype=int)
properties = regionprops(label_image, intensity_image=binary_global)
for region in properties:
if region['convex_area'] > area_thresh:
def overlay_labels(image, lbp, labels):
mask = np.logical_or.reduce([lbp == each for each in labels])
return label2rgb(mask, image=image, bg_label=0, alpha=0.5)
from skimage import io
from skimage import data, segmentation, color
from skimage.io import imread
from skimage import data
from skimage.future import graph
img = io.imread("../pants.jpg")
img_segments = segmentation.slic(img, compactness=30, n_segments=200)
out1 = color.label2rgb(img_segments, img, kind='avg')
segment_graph = graph.rag_mean_color(img, img_segments, mode='similarity')
img_cuts = graph.cut_normalized(img_segments, segment_graph)
normalized_cut_segments = color.label2rgb(img_cuts, img, kind='avg')
io.imshow(normalized_cut_segments)
io.show()
def preview(self, ips, para):
lab = segmentation.quickshift(ips.snap, para['ratio'],
para['kernel_size'], para['max_dist'],
para['sigma'])
ips.img[:] = color.label2rgb(lab, ips.snap, kind='avg')
def preview(self, ips, para):
lab = segmentation.felzenszwalb(ips.snap, para['scale'], para['sigma'],
para['min_size'])
ips.img[:] = color.label2rgb(lab, ips.snap, kind='avg')
def run(self, ips, snap, img, para=None):
lab = segmentation.slic(snap, para['n_segments'], para['compactness'],
para['max_iter'], para['sigma'])
return color.label2rgb(lab, snap, kind='avg')
from minisom import MiniSom
import numpy as np
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('input_image', type=str, help='input image path')
parser.add_argument('num_superpixel', type=int, help='number of segments')
parser.add_argument('compactness', type=int, help='compactness param of SLIC')
args = parser.parse_args()
#img = data.coffee()
img = io.imread(args.input_image)
labels = segmentation.slic(img,
n_segments=args.num_superpixel,
compactness=args.compactness)
out1 = color.label2rgb(labels, img, kind='avg')
io.imshow(out1)
io.show()
pixels = np.reshape(out1, (out1.shape[0] * out1.shape[1], 3)) / 255
print('training...')
som = MiniSom(2,
1,
3,
sigma=1.,
learning_rate=0.2,
neighborhood_function='bubble')
som.random_weights_init(pixels)
starting_weights = som.get_weights().copy() # saving the starting weights
import numpy as np
import matplotlib.pyplot as plt
from skimage import io, data, util, filters, color
from skimage.morphology import watershed
kitten = color.rgb2gray(io.imread("../images/kitten.jpeg"))
kitten_edge = filters.sobel(kitten) # use edge detection algo before watershed
grid = util.regular_grid(
kitten.shape, n_points=300) # find 300 points evenly spaced in the image
# The seed matrix is the same shape as the original image, and it contains integers in the range [1, size of image]
seeds = np.zeros(kitten.shape, dtype=int)
seeds[grid] = np.arange(seeds[grid].size).reshape(seeds[grid].shape) + 1
w0 = watershed(kitten_edge, seeds)
water_classic = color.label2rgb(w0, kitten, alpha=0.4, kind="overlay")
plt.figure(figsize=(8, 8))
plt.imshow(water_classic)
def prepareDataset(basedir='WeizmannSingleScale/horses/training/images/',
labeldir='WeizmannSingleScale/horses/training/masks/'):
global pixelClasses
dataSetX = []
dataSetX_layer2 = []
dataSetY = []
datasetGroundTruth = []
for (dirpath, dirnames, filenames) in walk(basedir):
n = 0
for imageFilename in filenames:
#if n>=1:
# break
print imageFilename
n = n + 1
# Read RGB and label image
image = img_as_float(skimageIO.imread(basedir + imageFilename))
bgrImage = cv2.imread(basedir + imageFilename, cv2.IMREAD_COLOR)
#bgrImage = cv2.fastNlMeansDenoisingColored(bgrImage)
#bgrImage = exposure.adjust_sigmoid(bgrImage)
labelImage = img_as_float(
skimageIO.imread(labeldir +
imageFilename.replace('image', 'mask')))
if len(image.shape) == 2:
image = color.gray2rgb(image)
# Resize
#image = resize(image,(120,120), preserve_range=True )
#bgrImage = resize(bgrImage,(120,120), preserve_range=True)
#labelImage = resize(labelImage,(120,120), preserve_range=True)
# Scan label image for additional classes
if len(labelImage.shape) == 2:
labelImageRGB = color.gray2rgb(labelImage)
else:
labelImageRGB = labelImage
#for i in range(0,labelImageRGB.shape[0]) :
# for j in range(0,labelImageRGB.shape[1]) :
# if len(pixelClasses) == 0 :
# pixelClasses.append(labelImageRGB[i][j])
# else :
# isAlreadyAPixelClass = False
# for pixelClass in pixelClasses:
# if numpy.array_equal(pixelClass, labelImageRGB[i][j]) :
# isAlreadyAPixelClass = True
# break
# if not isAlreadyAPixelClass :
# pixelClasses.append(labelImageRGB[i][j])
#print pixelClasses
# Derive superpixels and get their average RGB component
segments = slic(image, n_segments=500, sigma=1.0)
rgb_segments = img_as_ubyte(mark_boundaries(image, segments))
label_segments = img_as_ubyte(
mark_boundaries(labelImageRGB, segments))
avg_rgb = color.label2rgb(segments, image, kind='avg')
avg_label = color.label2rgb(segments, labelImageRGB, kind='avg')
#avg_cie_rgb = color.rgb2lab(avg_rgb)
#avg_cie_label = color.rgb2lab(avg_label)
# Create graph of superpixels and compute their centers
vertices, edges = make_graph(segments)
gridx, gridy = numpy.mgrid[:segments.shape[0], :segments.shape[1]]
centers = dict()
for v in vertices:
centers[v] = [
gridy[segments == v].mean(), gridx[segments == v].mean()
]
#print vertices
#print edges
#print centers
# Build training instances
n_features = []
edge_features = []
n_labels = []
# Compute image centers
centerX = labelImageRGB.shape[1] / 2.0
centerY = labelImageRGB.shape[0] / 2.0
for v in vertices:
# unary features layer 1 - average rgb of superpixel, histogram of patch surrounding center and CNN features
avg_rgb2 = avg_rgb[int(centers[v][1])][int(centers[v][0])]
hist, hogFeatures = getHistogramFeatures(bgrImage,
int(centers[v][1]),
int(centers[v][0]),
forUnaryFeature=True)
#relativeX = (centers[v][1] - centerX) / centerX
#relativeY = (centers[v][0] - centerY) / centerY
node_feature = numpy.concatenate([avg_rgb2, hist, hogFeatures])
n_features.append(node_feature)
# label
minEuclideanDistance = numpy.inf # simulate infinity
pixelClass = -1
for i in range(0, len(pixelClasses)):
# set the label of the superpixel to the pixelClass with minimum euclidean distance
dist = numpy.linalg.norm(
avg_label[int(centers[v][1])][int(centers[v][0])] -
pixelClasses[i])
if dist < minEuclideanDistance:
pixelClass = i
minEuclideanDistance = dist
n_labels.append(pixelClass)
histogramCache = {}
for e in edges:
# pairwise feature layer 1 - histogram distance, avg RGB euclidean distance of adjacent superpixels , texture similarity
dist = numpy.linalg.norm(
avg_rgb[int(centers[e[0]][1])][int(centers[e[0]][0])] -
avg_rgb[int(centers[e[1]][1])][int(centers[e[1]][0])])
if e[0] not in histogramCache:
hist1, lbphist1 = getHistogramFeatures(
bgrImage, int(centers[e[0]][1]), int(centers[e[0]][0]))
histogramCache[e[0]] = {'hist': hist1, 'lbphist': lbphist1}
else:
hist1 = histogramCache[e[0]]['hist']
lbphist1 = histogramCache[e[0]]['lbphist']
if e[1] not in histogramCache:
hist2, lbphist2 = getHistogramFeatures(
bgrImage, int(centers[e[1]][1]), int(centers[e[1]][0]))
histogramCache[e[1]] = {'hist': hist2, 'lbphist': lbphist2}
else:
hist2 = histogramCache[e[1]]['hist']
lbphist2 = histogramCache[e[1]]['lbphist']
histogramDist = cv2.compareHist(hist1, hist2,
3) # Bhattacharyya distance
textureSimilarity = kullback_leibler_divergence(
lbphist1, lbphist2) # KL divergence
pairwise_feature = numpy.array(
[dist, histogramDist, textureSimilarity])
edge_features.append(pairwise_feature)
# Add to dataset
dataSetX.append((numpy.array(n_features), numpy.array(edges),
numpy.array(edge_features)))
dataSetY.append(numpy.array(n_labels))
return dataSetX, dataSetY
def test_rrssd_box_coder_synthetic():
label_image = np.zeros((768, 768), dtype=np.uint8)
cv2.fillConvexPoly(
label_image,
np.expand_dims(cv2.boxPoints(((100, 100), (100, 20), 0)),
1).astype(int), (1, 1, 1))
cv2.fillConvexPoly(
label_image,
np.expand_dims(cv2.boxPoints(((200, 100), (100, 20), 45)),
1).astype(int), (2, 2, 2))
cv2.fillConvexPoly(
label_image,
np.expand_dims(cv2.boxPoints(((100, 200), (100, 20), 90)),
1).astype(int), (3, 3, 3))
cv2.fillConvexPoly(
label_image,
np.expand_dims(cv2.boxPoints(((200, 200), (100, 20), 135)),
1).astype(int), (4, 4, 4))
cv2.fillConvexPoly(
label_image,
np.expand_dims(cv2.boxPoints(((100 + 200, 100), (20, 100), 0)),
1).astype(int), (5, 5, 5))
cv2.fillConvexPoly(
label_image,
np.expand_dims(cv2.boxPoints(((200 + 200, 100), (20, 100), 45)),
1).astype(int), (6, 6, 6))
cv2.fillConvexPoly(
label_image,
np.expand_dims(cv2.boxPoints(((100 + 200, 200), (20, 100), 90)),
1).astype(int), (7, 7, 7))
cv2.fillConvexPoly(
label_image,
np.expand_dims(cv2.boxPoints(((200 + 200, 200), (20, 100), 135)),
1).astype(int), (8, 8, 8))
cv2.fillConvexPoly(
label_image,
np.expand_dims(cv2.boxPoints(((100, 105 + 200), (100, 20), 45. / 2)),
1).astype(int), (9, 9, 9))
cv2.fillConvexPoly(
label_image,
np.expand_dims(cv2.boxPoints(((200, 100 + 200), (16, 4), 49)),
1).astype(int), (10, 10, 10))
cv2.fillConvexPoly(
label_image,
np.expand_dims(cv2.boxPoints(((200, 100 + 210), (100, 20), 49)),
1).astype(int), (11, 11, 11))
cv2.fillConvexPoly(
label_image,
np.expand_dims(cv2.boxPoints(((200, 200 + 200), (100, 20), 165)),
1).astype(int), (12, 12, 12))
cv2.fillConvexPoly(
label_image,
np.expand_dims(cv2.boxPoints(((300, 300), (10, 6), 49)),
1).astype(int), (13, 13, 13))
cv2.fillConvexPoly(
label_image,
np.expand_dims(cv2.boxPoints(((300 + 50, 300), (200, 40), 90)),
1).astype(int), (14, 14, 14))
cv2.fillConvexPoly(
label_image,
np.expand_dims(cv2.boxPoints(((300 + 70, 300), (100, 20), 90)),
1).astype(int), (15, 15, 15))
cv2.fillConvexPoly(
label_image,
np.expand_dims(cv2.boxPoints(((500, 500), (2, 3), 9)), 1).astype(int),
(16, 16, 16))
cv2.fillConvexPoly(
label_image,
np.expand_dims(cv2.boxPoints(((510, 500), (2, 3), 19)), 1).astype(int),
(17, 17, 17))
cv2.fillConvexPoly(
label_image,
np.expand_dims(cv2.boxPoints(((540, 500), (4, 6), 29)), 1).astype(int),
(18, 18, 18))
cv2.fillConvexPoly(
label_image,
np.expand_dims(cv2.boxPoints(((560, 500), (2, 3), 39)), 1).astype(int),
(19, 19, 19))
image = (label2rgb(label_image, bg_label=0) * 255).astype(np.uint8)
# Test what happens if we rotate
# image = np.rot90(image).copy()
# label_image = np.rot90(label_image).copy()
rbboxes = instance_mask_to_rbboxes(label_image)
print(rbboxes)
labels = np.zeros(len(rbboxes), dtype=np.intp)
box_coder = RRNBoxCoder(768, 768)
loc_targets, cls_targets, anchors = box_coder.encode([], [],
return_anchors=True)
loc_targets, cls_targets, anchors = box_coder.encode(rbboxes,
labels,
return_anchors=True)
print(loc_targets.shape, cls_targets.shape, anchors.shape)
print('Object anchors', (cls_targets > 0).sum())
print('Background anchors', (cls_targets == 0).sum())
print('Ignore anchors', (cls_targets == -1).sum())
# cls_targets = np.expand_dims(cls_targets)
# print(cls_targets_one_hot.shape)
dec_boxes, dec_scores = box_coder.decode(loc_targets, cls_targets)
print(dec_boxes)
print('Total anchors', len(anchors))
for bbox in dec_boxes:
visualize_rbbox(image, bbox, (255, 0, 255), thickness=3)
for bbox in rbboxes:
visualize_rbbox(image, bbox, (0, 255, 0), thickness=1)
for bbox in anchors:
visualize_rbbox(image, bbox, (255, 255, 255), thickness=1)
cv2.imshow('overlays', image)
cv2.waitKey(-1)
