def test_rgba2rgb_conversion(self):
rgba = self.img_rgba
rgb = rgba2rgb(rgba)
expected = np.array([[[1, 1, 1],
[0, 0.5, 1],
[0.5, 0.75, 1]]]).astype(np.float)
self.assertEqual(rgb.shape, expected.shape)
assert_almost_equal(rgb, expected)
def punchhole_removal(im):
import numpy as np
from PIL import Image
from skimage import io
from skimage.color import rgba2rgb, rgb2gray
from skimage.transform import hough_circle, hough_circle_peaks
from skimage.feature import canny
from skimage.draw import circle
from skimage.util import img_as_ubyte
''' check for punch holes and remove '''
max_peaks = 24 #maximum number of peaks to be found. changed from 99 to 24 for reducing the unnecessary punch holes being filled.
img = np.array(im)# Load picture .
img_rgb = rgba2rgb(img)# convert to RGB
img_gray = rgb2gray(img_rgb)# convert to gray
image = img_as_ubyte(img_gray)
width, height = image.shape
x1 = punchhole_margin
x2 = (int)(width - punchhole_margin)
y1 = (int)(height - punchhole_margin)
y2 = punchhole_margin
edges = canny(image, 3, 10, 40) # perform canny to detect the edges
hough_radii = np.arange(31, 34, 1) #get the radius range with step as 1.
hough_res = hough_circle(edges, hough_radii) # detect the circles centres coordinates
# Select the most prominent circles based on the max_peaks
accums, cx, cy, radii = hough_circle_peaks(hough_res, hough_radii,total_num_peaks=max_peaks)
for center_y, center_x, radius in zip(cy, cx, radii):
#if the circles centres fall in the border regions,
#get the dominant color near the hole and fill the hole with a linear gradient of the dominant color
if(((0 < center_y < width) and (0 < center_x < y2)) or \
((0 < center_y < width) and (y1 < center_x < height)) or\
((0 < center_y < x1) and (0 < center_x < height)) or \
((x2 < center_y < width) and (0 < center_x < height))):
index=0
rr, cc= circle(center_y, center_x, radius+1, img.shape)
dominantpix = dominantcolor(center_x, center_y, radius, img)
dark_grad = [dominantpix[0], dominantpix[1],dominantpix[2]]
light_grad = [dominantpix[0]+1, dominantpix[1]+1, dominantpix[2]+1]
#white_grad = [255,255,255]
RGBA_list = lineargradient(dark_grad,light_grad,len(list(rr)))
for i , j in zip(list(rr), list(cc)):
pixlist = RGBA_list[index]
pixtuple = tuple(pixlist)
img[i,j]= (pixtuple[0], pixtuple[1], pixtuple[2], 255)
index += 1
finalimage=Image.fromarray(img)
return finalimage
def read_image(filename, number_of_channels):
""" read_image using skimage
The output is a 3-dim image [H, W, C]
"""
if number_of_channels == 1:
image = io.imread(filename, as_gray=True)
image = imgproc.toUINT8(image)
assert len(image.shape) == 2
image = np.expand_dims(image, axis=2) # H,W,C
assert len(image.shape) == 3 and image.shape[2] == 1
elif number_of_channels == 3:
image = io.imread(filename)
if len(image.shape) == 2:
image = color.gray2rgb(image)
elif image.shape[2] == 4:
image = color.rgba2rgb(image)
image = imgproc.toUINT8(image)
assert len(image.shape) == 3 and image.shape[2] == 3
else:
raise ValueError("number_of_channels must be 1 or 3")
if not os.path.exists(filename):
raise ValueError(filename + " does not exist!")
return image
def crf(original_image, annotated_image):
rd.seed(123)
if len(original_image.shape) < 3:
original_image = gray2rgb(original_image)
if len(original_image.shape) == 3 and original_image.shape[2] == 4:
original_image = rgba2rgb(original_image)
original_image = img_as_ubyte(original_image)
annotated_image = np.moveaxis(annotated_image, -1, 0)
annotated_image = annotated_image.copy(order='C')
d = dcrf.DenseCRF2D(
original_image.shape[1], original_image.shape[0], 3)
U = unary_from_softmax(annotated_image)
d.setUnaryEnergy(U)
d.addPairwiseGaussian(sxy=(3, 3), compat=3, kernel=dcrf.DIAG_KERNEL,
normalization=dcrf.NORMALIZE_SYMMETRIC)
d.addPairwiseBilateral(sxy=(80, 80), srgb=(13, 13, 13), rgbim=original_image,
compat=10,
kernel=dcrf.DIAG_KERNEL,
normalization=dcrf.NORMALIZE_SYMMETRIC)
Q = d.inference(NB_ITERATIONS)
MAP = np.argmax(Q, axis=0).reshape(
original_image.shape[0], original_image.shape[1])
result = np.zeros((MAP.shape[0], MAP.shape[1], 3))
result[:, :, 2] = MAP
result[:, :, 2][result[:, :, 2] == 2] = 4
result[:, :, 2][result[:, :, 2] == 1] = 2
result[:, :, 2][result[:, :, 2] == 0] = 1
return result
def __init__(self, fname, format=None, resize=None, as_gray=False,
electrodes=None, metadata=None, compress=False):
img = imread(fname, format=format)
# Build the metadata container:
if metadata is None:
metadata = {}
metadata['source'] = fname
metadata['source_shape'] = img.shape
# Convert to grayscale if necessary:
if as_gray:
if img.shape[-1] == 4:
# Convert the transparent background to black:
img = rgba2rgb(img, background=(0, 0, 0))
img = rgb2gray(img)
# Resize if necessary:
if resize is not None:
img = img_resize(img, resize)
# Store the original image shape for resizing and color conversion:
self.img_shape = img.shape
# Convert to float array in [0, 1] and call the Stimulus constructor:
super(ImageStimulus, self).__init__(img_as_float(img).ravel(),
time=None, electrodes=electrodes,
metadata=metadata,
compress=compress)
def rgb_to_od(img: PIL.Image.Image) -> PIL.Image.Image:
"""Convert from RGB to optical density (OD_RGB) space.
Parameters
----------
img : PIL.Image.Image
Input image
Returns
-------
PIL.Image.Image
Image in OD space
"""
if img.mode == "RGBA":
img_arr = np.array(sk_color.rgba2rgb(img))
warn("Input image must be RGB. "
"NOTE: the image will be converted to RGB before HED conversion.")
else:
img_arr = np.array(img)
mask = img_arr == 0
img_arr[mask] = 1
od_arr = np.maximum(-1 * np.log10(img_arr / 255), 1e-6).round(10)
od_arr = sk_exposure.rescale_intensity(od_arr)
return np_to_pil(od_arr)
def test_lab_encode():
img = read_image(test_image)
img = transform.resize(img, (64, 64))
img = color.rgba2rgb(img)
img = convert_rgb_to_lab(img)
# plot_image(img)
start = time.time()
encoded = soft_encode_lab_img(img)
for row in range(encoded.shape[0]):
for col in range(encoded.shape[1]):
dist = encoded[row, col, :]
if not np.isclose(1, dist.sum()):
print(dist)
print(dist.sum())
exit()
end = time.time()
total = end - start
print("encoding took", total, "seconds")
def felzen_segment():
global segments, orig
img_b64 = request.values['data'].split("base64,")[1]
val = 600 - int(float(request.values['val']))
image_result = open('orig.png', 'wb')
image_result.write(base64.b64decode(img_b64))
img = io.imread('orig.png')
img = color.rgba2rgb(img)
segments = segmentation.felzenszwalb(img, min_size=val)
segmented_img = segmentation.mark_boundaries(img, segments)
back_img = np.zeros((segmented_img.shape[0], segmented_img.shape[1], 4))
back_img[:, :, 0] = img[:, :, 0] * 255
back_img[:, :, 1] = img[:, :, 1] * 255
back_img[:, :, 2] = img[:, :, 2] * 255
back_img[:, :, 3] = 255
orig = img
for file in os.listdir("./masks/"):
os.remove("./masks/" + file)
plt.imsave("masks/back.png", back_img.astype(np.uint8))
plt.imsave("segment.jpg", segmented_img)
np.save("segments", segments)
print('Image received: {}'.format(img.shape))
response = jsonify({'message': 'Happy Noises'})
return response
def test_rgba2rgb_dtype(self):
rgba = self.img_rgba.astype('float64')
rgba32 = img_as_float32(rgba)
assert rgba2rgb(rgba).dtype == rgba.dtype
assert rgba2rgb(rgba32).dtype == rgba32.dtype
def main():
list_of_colors = read_color_codes("classes.csv")
image_names_list = os.listdir(in_dir_images)
image_index = 3 # 18
target_index = 42
img = io.imread(os.path.join(in_dir_images, image_names_list[image_index]))
print("IMG --> ", image_names_list[image_index])
print("CODED TARGET --> ", target_index + 1)
if img.shape[2] == 4:
img = color.rgba2rgb(img)
elif img.shape[2] < 3 or img.shape[2] > 5:
print("Our img non RGB or RGBA. We don't know, what to do...")
return 1
mask_coded_target = np.all(img == list_of_colors[target_index], axis=-1)
img_edges = feature.canny(mask_coded_target)
# Проходим по строкам
line_1_points = list()
line_1_xs = list()
line_1_ys = list()
epsilon = 2
for row in range(img_edges.shape[0]):
row_edges = np.nonzero(img_edges[row])
row_edges = row_edges[0]
if len(row_edges):
avr_in_row = (row_edges[0] + row_edges[-1]) / 2
if row <= 0 + epsilon or row >= img_edges.shape[0] - epsilon:
print("BAD ROW. CODED TARGET --> ", target_index)
line_1_points.clear()
line_1_xs.clear()
line_1_ys.clear()
break
line_1_points.append([avr_in_row, row])
line_1_xs.append(avr_in_row)
line_1_ys.append(row)
# Проходим по столбцам
line_2_points = list()
line_2_xs = list()
line_2_ys = list()
epsilon = 2
for col in range(img_edges.shape[1]):
col_edges = np.nonzero(img_edges[:, col])
col_edges = col_edges[0]
if len(col_edges):
avr_in_col = (col_edges[0] + col_edges[-1]) / 2
if col <= 0 + epsilon or col >= img_edges.shape[1] - epsilon:
print("BAD COL. CODED TARGET --> ", target_index)
line_2_points.clear()
line_2_xs.clear()
line_2_ys.clear()
break
line_2_points.append([col, avr_in_col])
line_2_xs.append(col)
line_2_ys.append(avr_in_col)
# находим пересечение двух прямых
xs = None
ys = None
if len(line_1_xs) * len(line_1_ys) and len(line_2_xs) * len(line_2_ys):
m1, b1 = np.polyfit(line_1_xs, line_1_ys, 1)
m2, b2 = np.polyfit(line_2_xs, line_2_ys, 1)
A = np.array([[m1, -1], [m2, -1]])
B = np.array([[-b1], [-b2]])
xs, ys = np.linalg.solve(A, B)
print("xs = ", xs, "ys = ", ys)
fig, (ax1, ax2) = plt.subplots(nrows=1,
ncols=2,
figsize=(8, 3),
sharex=True,
sharey=True)
ax1.imshow(img)
ax1.axis("off")
ax1.set_title("Image")
ax2.imshow(img_edges)
# Отображаем точки первой линии
for point in line_1_points:
plt.scatter(point[0], point[1], s=1, facecolor='red')
# Отображаем точки второй линии
for point in line_2_points:
plt.scatter(point[0], point[1], s=1, facecolor='red')
if xs is not None and ys is not None:
plt.scatter(xs, ys, s=2, facecolor='green')
ax2.axis("off")
ax2.set_title("Edges")
fig.tight_layout()
plt.show()
def recolor(request):
if request.method == 'POST':
json_name = request.POST['case']
banner = getImageFromJson(os.path.join('./case' , json_name + '.json'))
#banner = getImageFromJson('0073_1760.json')
target = banner.paint()
target.save('./static/img/target.png')
palette_gen = request.POST['paletteGen']
render_op = request.POST['render']
# iseva = request.POST['iseva']
keys = request.POST.keys()
iseva = False
if 'iseva' in keys:
iseva = True
# assert False
orig_score = banner.evaluate()
# change background and text
childrens = banner.childrens
background = childrens[0]
element = background
for item in childrens:
if item.category == 'background':
background = item
elif item.category == 'element':
element = item
url_elem = 'http:' + element.src
elem = io.imread(url_elem)
mask = np.ones(elem[:,:,0].shape) * 255
if elem.shape[2] == 4:
mask = elem[:,:,3]
elem = color.rgba2rgb(elem)
elem = img_as_ubyte(elem)
elem_color = getProductColor(elem , mask)
print elem_color
elem_palette_img = getImagePalette(elem_color)
scipy.misc.imsave('./static/img/elem_palette.jpg' , elem_palette_img)
pal = palettes.Palette(elem_color)
if palette_gen == '0':
palette = pal.getRecColorFromColormind()
else:
pass
# palette = [[49,47,49],[91,83,81],[133,155,143],[226,209,167],[235,198,126]]
score = pal.palette_score(palette)
scipy.misc.imsave('./static/img/palette.jpg' , getImagePalette(palette))
contrasts = []
order = []
scores = []
for item in permutations(range(2 , 5) , 2):
index_bak , index_text = item
bak_l , _ , _ = colormap.rgb2lab(palette[index_bak][0] , palette[index_bak][1] , palette[index_bak][2])
text_l , _, _ = colormap.rgb2lab(palette[index_text][0] , palette[index_text][1] , palette[index_text][2])
contrast = abs(bak_l - text_l)
contrasts.append(contrast)
#order.append(item)
banner.childChange(bak_change(background , palette[index_bak:index_bak + 1] , render_op))
for item in childrens:
print item.category
print item.filename
if 'title' in item.category:
# banner.childChange(title_change(item , palette[index_text:index_text + 1]))
# print 'adasd'
if color_select(item):
banner.childChange(title_change(item , palette[index_text:index_text + 1]))
else:
banner.childChange(bak_change(item , palette[index_text:index_text + 1] , render_op))
img = banner.paint()
if iseva:
scores.append(banner.evaluate())
img.save(os.path.join('./static/img/' , str(index_bak) + '_' + str(index_text) + '.png'))
order.append(str(index_bak) + '_' + str(index_text) + '.png')
print contrasts
print order
# print score , scores
print score
if iseva:
res = zip(order , scores)
else:
res = order
return render_to_response('index.html' , {'flag':True , 'pics':res , 'score':score , 'iseval':iseva , 'orgin_score':orig_score})
self.r1 = min(mask[:, 0])
self.r2 = max(mask[:, 0])
self.c1 = min(mask[:, 1])
self.c2 = max(mask[:, 1])
self.data = np.ones((28, 28), dtype=np.float64)
for bb in mask:
print(type(o[bb[0], bb[1]]))
self.data[bb[0] - self.r1 + 4, bb[1] - self.c1 + 4] = o[bb[0],
bb[1]]
self.pixel = 255 - img_as_ubyte(self.data)
self.pixel = np.array(self.pixel).reshape(28 * 28)
filename = os.path.join('4.png')
moon = io.imread(filename)
moon = rgb2gray(rgba2rgb(moon))
block_size = 11
local_thresh = threshold_local(moon, block_size, offset=0.001)
binary = moon >= local_thresh
a = np.argwhere(binary == False)
kmeans = KMeans(n_clusters=4, random_state=0).fit(a)
b = []
for i in range(4):
b.append(figure(a[kmeans.predict(a) == i], moon))
b = sorted(b, key=operator.attrgetter('c1'))
im = Image.fromarray(img_as_ubyte(b[0].data))
im.save("./figures/1.jpg")
im = Image.fromarray(img_as_ubyte(b[1].data))
im.save("./figures/2.png")
im = Image.fromarray(img_as_ubyte(b[2].data))
def wmcnn_hdf5(path, full=False):
def expanddims(array, expand=2):
for i in range(expand):
array = np.expand_dims(array, 0)
return array
# from matplotlib import pyplot as plt
# from skimage import measure
scale = 2
if not full:
size_label = 96
stride = 48
else:
size_label = 400
size_input = size_label / scale
batch_size = 400
classes = 7
# downsizes = [1, 0.7, 0.5]
id_files = join(path, 'id_files.txt')
# id_files = join(path, 'test_id.txt')
with open(id_files, 'r') as f:
article = f.readlines()
fnum = len(article) // classes
order = np.random.permutation(fnum)
training = order[:int(fnum * 0.7)]
testing = order[len(training):]
img_list = []
for i, line in enumerate(article):
img = io.imread(path + line[2:].strip('\n'))
if i % fnum in testing:
cls = line.split('/')[1]
mkdir_if_not_exist(path + 'test/' + cls)
io.imsave(path + 'test/' + line[2:].strip('\n'), img)
else:
if img.shape[2] == 4:
img = color.rgba2rgb(img)
img_ycbcr = color.rgb2ycbcr(img) / 255
(rows, cols, channel) = img_ycbcr.shape
img_y, img_cb, img_cr = np.split(img_ycbcr,
indices_or_sections=channel,
axis=2)
img_list.append(img_y)
for idx, img_gt in enumerate(img_list):
if idx == 0:
hf = h5py.File('D:/data/rsscn7/full_wdata.h5', 'w')
d_data = hf.create_dataset("data",
(batch_size, 1, size_input, size_input),
maxshape=(None, 1, size_input,
size_input),
dtype='float32')
d_ca = hf.create_dataset("CA",
(batch_size, 1, size_input, size_input),
maxshape=(None, 1, size_input,
size_input),
dtype='float32')
d_ch = hf.create_dataset("CH",
(batch_size, 1, size_input, size_input),
maxshape=(None, 1, size_input,
size_input),
dtype='float32')
d_cv = hf.create_dataset("CV",
(batch_size, 1, size_input, size_input),
maxshape=(None, 1, size_input,
size_input),
dtype='float32')
d_cd = hf.create_dataset("CD",
(batch_size, 1, size_input, size_input),
maxshape=(None, 1, size_input,
size_input),
dtype='float32')
else:
hf = h5py.File('D:/data/rsscn7/full_wdata.h5', 'a')
d_data = hf['data']
d_ca = hf['CA']
d_ch = hf['CH']
d_cv = hf['CV']
d_cd = hf['CD']
count = 0
if not full:
d_data.resize([idx * batch_size + 392, 1, size_input, size_input])
d_ca.resize([idx * batch_size + 392, 1, size_input, size_input])
d_ch.resize([idx * batch_size + 392, 1, size_input, size_input])
d_cv.resize([idx * batch_size + 392, 1, size_input, size_input])
d_cd.resize([idx * batch_size + 392, 1, size_input, size_input])
for flip in range(2):
for degree in range(4):
# for downsize in downsizes:
img = img_gt.squeeze()
if flip == 1:
img = np.fliplr(img)
for turn in range(degree):
img = np.rot90(img)
# img = imresize(img, scalar_scale=downsize)
hei, wid = img.shape
# fig = plt.figure(figsize=(6, 3))
for x in range(0, hei - size_label, stride):
for y in range(0, wid - size_label, stride):
subim_label = img[x:x + size_label,
y:y + size_label]
subim_data = imresize(subim_label,
scalar_scale=1 / scale)
coeffs2 = pywt.dwt2(subim_label, 'bior1.1')
LL, (LH, HL, HH) = coeffs2
d_data[idx * batch_size + count] = expanddims(
subim_data, expand=2)
d_ca[idx * batch_size + count] = expanddims(
LL, expand=2)
d_ch[idx * batch_size + count] = expanddims(
LH, expand=2)
d_cv[idx * batch_size + count] = expanddims(
HL, expand=2)
d_cd[idx * batch_size + count] = expanddims(
HH, expand=2)
count += 1
else:
d_data.resize([idx * batch_size + 1, 1, size_input, size_input])
d_ca.resize([idx * batch_size + 1, 1, size_input, size_input])
d_ch.resize([idx * batch_size + 1, 1, size_input, size_input])
d_cv.resize([idx * batch_size + 1, 1, size_input, size_input])
d_cd.resize([idx * batch_size + 1, 1, size_input, size_input])
img = img_gt.squeeze()
im_data = imresize(img, scalar_scale=1 / scale)
coeffs2 = pywt.dwt2(img, 'bior1.1')
LL, (LH, HL, HH) = coeffs2
d_data[idx * batch_size + count] = expanddims(im_data, expand=2)
d_ca[idx * batch_size + count] = expanddims(LL, expand=2)
d_ch[idx * batch_size + count] = expanddims(LH, expand=2)
d_cv[idx * batch_size + count] = expanddims(HL, expand=2)
d_cd[idx * batch_size + count] = expanddims(HH, expand=2)
count += 1
batch_size = count
hf.close()
def GetPoseandCostsF(
cfg,
dlc_cfg,
sess,
inputs,
outputs,
cap,
nframes,
batchsize,
shelf_path,
):
"""Batchwise prediction of pose"""
strwidth = int(np.ceil(np.log10(nframes))) # width for strings
batch_ind = 0 # keeps track of which image within a batch should be written to
batch_num = 0 # keeps track of which batch you are at
if cfg["cropping"]:
cap.set_bbox(cfg["x1"], cfg["x2"], cfg["y1"], cfg["y2"])
nx, ny = cap.dimensions
frames = np.empty((batchsize, ny, nx, 3),
dtype="ubyte") # this keeps all frames in a batch
pbar = tqdm(total=nframes)
counter = 0
inds = []
if shelf_path:
db = shelve.open(
shelf_path,
protocol=pickle.DEFAULT_PROTOCOL,
)
else:
db = dict()
db["metadata"] = {
"nms radius":
dlc_cfg["nmsradius"],
"minimal confidence":
dlc_cfg["minconfidence"],
"sigma":
dlc_cfg.get("sigma", 1),
"PAFgraph":
dlc_cfg["partaffinityfield_graph"],
"PAFinds":
dlc_cfg.get("paf_best",
np.arange(len(dlc_cfg["partaffinityfield_graph"]))),
"all_joints": [[i] for i in range(len(dlc_cfg["all_joints"]))],
"all_joints_names": [
dlc_cfg["all_joints_names"][i]
for i in range(len(dlc_cfg["all_joints"]))
],
"nframes":
nframes,
}
while cap.video.isOpened():
frame = cap.read_frame(crop=cfg["cropping"])
key = "frame" + str(counter).zfill(strwidth)
if frame is not None:
# Avoid overwriting data already on the shelf
if isinstance(db, shelve.Shelf) and key in db:
continue
frame = img_as_ubyte(frame)
if frame.shape[-1] == 4:
frame = rgba2rgb(frame)
frames[batch_ind] = frame
inds.append(counter)
if batch_ind == batchsize - 1:
D = predict.predict_batched_peaks_and_costs(
dlc_cfg,
frames,
sess,
inputs,
outputs,
)
for ind, data in zip(inds, D):
db["frame" + str(ind).zfill(strwidth)] = data
del D
batch_ind = 0
inds.clear()
batch_num += 1
else:
batch_ind += 1
elif counter >= nframes:
if batch_ind > 0:
D = predict.predict_batched_peaks_and_costs(
dlc_cfg,
frames,
sess,
inputs,
outputs,
)
for ind, data in zip(inds, D):
db["frame" + str(ind).zfill(strwidth)] = data
del D
break
counter += 1
pbar.update(1)
cap.close()
pbar.close()
try:
db.close()
except AttributeError:
pass
return db, nframes
def get_volume_main(filepath, plate_color, ref_len, shape_type, additional_info = 0, debug = False, showing = False):
"""
shape type:
1: cube (e.g. cake)
2: ball (e.g. apple)
3: half-ball (e.g. bun)
4: cone (e.g. fried rice in the plate)
5: fixed-height (e.g. pizza)
6: irregular but nearly fixed shape (e.g. banana)
additional_info:
height, for type 5
volume per unit area, for type 6
"""
image_rgb = io.imread(filepath)
if (image_rgb.shape[2] == 4):
image_rgb = color.rgba2rgb(image_rgb)
image_rgb = transform.resize(image_rgb, (int(100*image_rgb.shape[0]/image_rgb.shape[1]), 100))
food, plate_long, plate_short = get_food_region(image_rgb, plate_color)
if showing:
io.imsave('original_image.jpg', image_rgb)
if debug:
f, ((ax0, ax1, ax2, ax3), (ax4, ax5, ax6, ax7)) = plt.subplots(ncols=4, nrows=2, figsize=(22, 8))
ax0.set_title('food')
ax0.imshow(image_rgb)
if shape_type == 1:
labels, labels2 = segment_food (image_rgb, food)
area, height = get_height_and_area(labels2)
volume = cal_volume_1(plate_long, plate_short, ref_len, area, height)
if debug:
ax2.set_title('segment')
ax2.imshow(labels)
ax3.set_title('segment2')
ax3.imshow(labels2)
if shape_type == 2:
volume = cal_volume_2(plate_long, plate_short, ref_len, food)
if shape_type == 3:
volume = cal_volume_3(plate_long, plate_short, ref_len, food)
if shape_type == 4:
volume = cal_volume_4(plate_long, plate_short, ref_len, food)
if shape_type == 5:
volume = cal_volume_5(plate_long, plate_short, ref_len, food, additional_info)
if shape_type == 6:
volume = cal_volume_6(plate_long, plate_short, ref_len, food, additional_info)
if debug:
print('The estimated volume is', volume, 'cm^3.\n(Plate size:', ref_len, 'cm; type of shape: #', shape_type, '.)')
for i in range(0, image_rgb.shape[0]):
for j in range(0, image_rgb.shape[1]):
if (food[i][j] == 0):
image_rgb[i][j] = [0,0,0]
ax1.set_title('food')
ax1.imshow(image_rgb)
if showing:
if shape_type == 1:
io.imsave('mid_result.jpg', labels2)
else:
io.imsave('mid_result.jpg', food)
return volume
plt.axis('off')
plt.title('orginal', fontsize = 12)
location = [2, 3, 4, 6, 7, 8]
for inter in range(6):
time1 = time.time()*1000
I1 = resize(img, (height*factor, length*factor, 3), order = inter)
time2 = time.time()*1000
diff = time2 - time1
plt.subplot(2, 4, location[inter])
plt.imshow(I1)
plt.axis('off')
plt.title('order: %i\n time: %.2fs' %(inter, diff), fontsize = 10)
I = imread('Images/tree.png')
I = rgba2rgb(I)
img = resize(I, (np.floor(0.2*I.shape[0]), np.floor(0.2*I.shape[1]), 3))
enlarge(img)
# The order terms are different methods used to extrapolate the colors of the new added pixels in the resizing. the order terms correspond to
#
# 0 = Nearest-neighbor,
# 1 = Bi-linear,
# 2 = Bi-quadratic,
# 3 = Bi-cubic,
# 4 = Bi-quartic,
# 5 = Bi-quintic,
#
# The time complexity increases with the order term respectively
async def tex(self, ctx: Context, *message: clean_content):
await ctx.trigger_typing()
print(message)
# Input filtering
if not message:
await ctx.send("Your message contained nothing to render")
if message[0] == "```tex":
message = ("```", *message[1:])
combined = " ".join([x.lstrip("@") for x in message])
if combined[0] != "`" or combined[-1] != "`":
await ctx.send("Please place your input in an inline code block")
return
tex_code = combined.lstrip("`").rstrip("`")
# May want to consider enforcing the $message$ requirement here
# May also want to disallow/convert $$message$$ environments here
# Matplotlib preamble
plt.clf()
plt.rc("text", usetex=True)
plt.rc("text.latex", preamble=r"\usepackage{amsmath}")
plt.rc("font", **{
"family": "serif",
"serif": ["Palatino"],
"size": 16
})
plt.axis("off")
# Generate the filename
filename = (tex_code + "-" +
str(hex(int(datetime.utcnow().timestamp()))).lstrip("0x") +
".png").replace(" ", "")
path_png = "{path}/{filename}".format(
path=CONFIG["FIG_SAVE_PATH"].rstrip("/"), filename=filename)
path_jpg = path_png.replace(".png", ".jpg")
try:
# Plot the latex and save it.
plt.text(0, 1, tex_code, color="white")
plt.savefig(path_png,
dpi=300,
bbox_inches="tight",
transparent=True)
except RuntimeError as r:
# Failed to render latex. Report error
print(r)
await ctx.send(
"Unable to render LaTeX. Please check that it's correct")
else:
# Generate a mask of the transparent regions in the image
img_arr = img_as_float(io.imread(path_png))
transparent_mask = np.array([1, 1, 1, 0])
img_mask = np.abs(img_arr - transparent_mask).sum(axis=2) < 1
# Generate the bounding box for the mask
mask_coords = np.array(np.nonzero(~img_mask))
top_left = np.min(mask_coords, axis=1) - [15, 15]
bottom_right = np.max(mask_coords, axis=1) + [15, 15]
# Crop the image and add a background layer
img_cropped = img_arr[top_left[0]:bottom_right[0],
top_left[1]:bottom_right[1]]
img_cropped = color.rgba2rgb(img_cropped,
background=IMAGE_BACKGROUND)
# Save the image, delete the PNG and set the permissions for the JPEG
io.imsave(path_jpg, img_cropped, quality=100)
os.chmod(path_jpg, 0o644)
os.remove(path_png)
# Load the image as a file to be attached to an image
img_file = File(path_jpg, filename="tex_output.jpg")
display_name = get_name_string(ctx.message)
await ctx.send(f"Here you go, {display_name}! :abacus:",
file=img_file)
def create_bags(self, dir_list):
bag_list = []
labels_list = []
for dir in dir_list:
# Get image name
img_name = os.path.basename(dir)
# bmp to pillow
img_dir = os.path.join(dir, img_name + '.bmp')
img = io.imread(img_dir)
if img.shape[2] == 4:
img = color.rgba2rgb(img)
if self.location_info:
xs = np.arange(0, 500)
xs = np.asarray([xs for i in range(500)])
ys = xs.transpose()
img = np.dstack((img, xs, ys))
# crop malignant cells
dir_epithelial = os.path.join(dir, img_name + '_epithelial.mat')
with open(dir_epithelial, 'rb') as f:
mat_epithelial = scipy.io.loadmat(f)
cropped_cells_epithelial = []
for (x, y) in mat_epithelial['detection']:
x = np.round(x)
y = np.round(y)
if self.data_augmentation:
x = x + np.round(np.random.normal(0, 3, 1))
y = y + np.round(np.random.normal(0, 3, 1))
if x < 13:
x_start = 0
x_end = 27
elif x > 500 - 14:
x_start = 500 - 27
x_end = 500
else:
x_start = x - 13
x_end = x + 14
if y < 13:
y_start = 0
y_end = 27
elif y > 500 - 14:
y_start = 500 - 27
y_end = 500
else:
y_start = y - 13
y_end = y + 14
cropped_cells_epithelial.append(img[int(y_start):int(y_end), int(x_start):int(x_end)])
# crop all other cells
dir_inflammatory = os.path.join(dir, img_name + '_inflammatory.mat')
dir_fibroblast = os.path.join(dir, img_name + '_fibroblast.mat')
dir_others = os.path.join(dir, img_name + '_others.mat')
with open(dir_inflammatory, 'rb') as f:
mat_inflammatory = scipy.io.loadmat(f)
with open(dir_fibroblast, 'rb') as f:
mat_fibroblast = scipy.io.loadmat(f)
with open(dir_others, 'rb') as f:
mat_others = scipy.io.loadmat(f)
all_coordinates = np.concatenate(
(mat_inflammatory['detection'], mat_fibroblast['detection'], mat_others['detection']), axis=0)
cropped_cells_others = []
for (x, y) in all_coordinates:
x = np.round(x)
y = np.round(y)
if self.data_augmentation:
x = x + np.round(np.random.normal(0, 3, 1))
y = y + np.round(np.random.normal(0, 3, 1))
if x < 13:
x_start = 0
x_end = 27
elif x > 500 - 14:
x_start = 500 - 27
x_end = 500
else:
x_start = x - 13
x_end = x + 14
if y < 13:
y_start = 0
y_end = 27
elif y > 500 - 14:
y_start = 500 - 27
y_end = 500
else:
y_start = y - 13
y_end = y + 14
cropped_cells_others.append(img[int(y_start):int(y_end), int(x_start):int(x_end)])
# generate bag
bag = cropped_cells_epithelial + cropped_cells_others
# store single cell labels
labels = np.concatenate((np.ones(len(cropped_cells_epithelial)), np.zeros(len(cropped_cells_others))),
axis=0)
# shuffle
if self.shuffle_bag:
zip_bag_labels = list(zip(bag, labels))
random.shuffle(zip_bag_labels)
bag, labels = zip(*zip_bag_labels)
# append every bag two times if training
if self.train:
for _ in [0, 1]:
bag_list.append(bag)
labels_list.append(labels)
else:
bag_list.append(bag)
labels_list.append(labels)
return bag_list, labels_list
def GetPoseandCostsF_from_assemblies(
cfg,
dlc_cfg,
sess,
inputs,
outputs,
cap,
nframes,
batchsize,
assemblies,
feature_dict,
extra_dict,
):
"""Batchwise prediction of pose"""
strwidth = int(np.ceil(np.log10(nframes))) # width for strings
batch_ind = 0 # keeps track of which image within a batch should be written to
batch_num = 0 # keeps track of which batch you are at
if cfg["cropping"]:
cap.set_bbox(cfg["x1"], cfg["x2"], cfg["y1"], cfg["y2"])
nx, ny = cap.dimensions
frames = np.empty((batchsize, ny, nx, 3),
dtype="ubyte") # this keeps all frames in a batch
pbar = tqdm(total=nframes)
counter = 0
inds = []
PredicteData = {}
while cap.video.isOpened():
frame = cap.read_frame(crop=cfg["cropping"])
key = "frame" + str(counter).zfill(strwidth)
if frame is not None:
# Avoid overwriting data already on the shelf
if key in feature_dict:
continue
frame = img_as_ubyte(frame)
if frame.shape[-1] == 4:
frame = rgba2rgb(frame)
frames[batch_ind] = frame
inds.append(counter)
if batch_ind == batchsize - 1:
preds = predict.predict_batched_peaks_and_costs(
dlc_cfg,
frames,
sess,
inputs,
outputs,
extra_dict=extra_dict)
if not preds:
continue
D, features = preds
for i, (ind, data) in enumerate(zip(inds, D)):
PredicteData["frame" + str(ind).zfill(strwidth)] = data
raw_coords = assemblies.get(ind)
if raw_coords is None:
continue
fname = "frame" + str(ind).zfill(strwidth)
feature_dict[fname] = _get_features_dict(
raw_coords,
features[i],
dlc_cfg["stride"],
)
batch_ind = 0
inds.clear()
batch_num += 1
else:
batch_ind += 1
elif counter >= nframes:
if batch_ind > 0:
preds = predict.predict_batched_peaks_and_costs(
dlc_cfg,
frames,
sess,
inputs,
outputs,
extra_dict=extra_dict)
if not preds:
continue
D, features = preds
for i, (ind, data) in enumerate(zip(inds, D)):
PredicteData["frame" + str(ind).zfill(strwidth)] = data
raw_coords = assemblies.get(ind)
if raw_coords is None:
continue
fname = "frame" + str(ind).zfill(strwidth)
feature_dict[fname] = _get_features_dict(
raw_coords,
features[i],
dlc_cfg["stride"],
)
break
counter += 1
pbar.update(1)
cap.close()
pbar.close()
feature_dict.close()
PredicteData["metadata"] = {
"nms radius":
dlc_cfg["nmsradius"],
"minimal confidence":
dlc_cfg["minconfidence"],
"sigma":
dlc_cfg.get("sigma", 1),
"PAFgraph":
dlc_cfg["partaffinityfield_graph"],
"PAFinds":
dlc_cfg.get("paf_best",
np.arange(len(dlc_cfg["partaffinityfield_graph"]))),
"all_joints": [[i] for i in range(len(dlc_cfg["all_joints"]))],
"all_joints_names": [
dlc_cfg["all_joints_names"][i]
for i in range(len(dlc_cfg["all_joints"]))
],
"nframes":
nframes,
}
return PredicteData, nframes
def create_bags_one_type(self, dir_list):
"""Create bags containing only one type of nucleus."""
bag_list = []
labels_list = []
for dir in dir_list:
# Get image name
img_name = dir.split('/')[-1]
# bmp to pillow
img_dir = dir + '/' + img_name + '.bmp'
img = io.imread(img_dir)
if img.shape[2] == 4:
img = color.rgba2rgb(img)
if self.location_info:
xs = np.arange(0, 500)
xs = np.asarray([xs for i in range(500)])
ys = xs.transpose()
img = np.dstack((img, xs, ys))
# crop nucleus_type cells
dir_nucleus_type = dir + '/' + img_name + '_' + self.nucleus_type + '.mat'
with open(dir_nucleus_type, 'rb') as f:
mat_nucleus_type = scipy.io.loadmat(f)
cropped_cells = []
for (x, y) in mat_nucleus_type['detection']:
x = np.round(x)
y = np.round(y)
if self.data_augmentation:
x = x + np.round(np.random.normal(0, 3, 1))
y = y + np.round(np.random.normal(0, 3, 1))
if x < 13:
x_start = 0
x_end = 27
elif x > 500 - 14:
x_start = 500 - 27
x_end = 500
else:
x_start = x - 13
x_end = x + 14
if y < 13:
y_start = 0
y_end = 27
elif y > 500 - 14:
y_start = 500 - 27
y_end = 500
else:
y_start = y - 13
y_end = y + 14
cropped_cells.append(img[int(y_start):int(y_end), int(x_start):int(x_end)])
# if image doesn't contain any specific type nucleus, move to the next image
if cropped_cells == []:
continue
# generate bag
bag = cropped_cells
# store single cell labels
if self.nucleus_type == 'epithelial':
labels = np.ones(len(cropped_cells))
else:
labels = np.zeros(len(cropped_cells))
# shuffle
if self.shuffle_bag:
zip_bag_labels = list(zip(bag, labels))
random.shuffle(zip_bag_labels)
bag, labels = zip(*zip_bag_labels)
# append every bag two times if training
if self.train:
for _ in [0, 1]:
bag_list.append(bag)
labels_list.append(labels)
else:
bag_list.append(bag)
labels_list.append(labels)
# bag_list.append(bag)
# labels_list.append(labels)
return bag_list, labels_list
def GetPoseandCostsS(cfg, dlc_cfg, sess, inputs, outputs, cap, nframes,
shelf_path):
"""Non batch wise pose estimation for video cap."""
strwidth = int(np.ceil(np.log10(nframes))) # width for strings
if cfg["cropping"]:
cap.set_bbox(cfg["x1"], cfg["x2"], cfg["y1"], cfg["y2"])
if shelf_path:
db = shelve.open(
shelf_path,
protocol=pickle.DEFAULT_PROTOCOL,
)
else:
db = dict()
db["metadata"] = {
"nms radius":
dlc_cfg["nmsradius"],
"minimal confidence":
dlc_cfg["minconfidence"],
"sigma":
dlc_cfg.get("sigma", 1),
"PAFgraph":
dlc_cfg["partaffinityfield_graph"],
"PAFinds":
dlc_cfg.get("paf_best",
np.arange(len(dlc_cfg["partaffinityfield_graph"]))),
"all_joints": [[i] for i in range(len(dlc_cfg["all_joints"]))],
"all_joints_names": [
dlc_cfg["all_joints_names"][i]
for i in range(len(dlc_cfg["all_joints"]))
],
"nframes":
nframes,
}
pbar = tqdm(total=nframes)
counter = 0
while cap.video.isOpened():
frame = cap.read_frame(crop=cfg["cropping"])
key = "frame" + str(counter).zfill(strwidth)
if frame is not None:
# Avoid overwriting data already on the shelf
if isinstance(db, shelve.Shelf) and key in db:
continue
frame = img_as_ubyte(frame)
if frame.shape[-1] == 4:
frame = rgba2rgb(frame)
dets = predict.predict_batched_peaks_and_costs(
dlc_cfg,
np.expand_dims(frame, axis=0),
sess,
inputs,
outputs,
)
db[key] = dets[0]
del dets
elif counter >= nframes:
break
counter += 1
pbar.update(1)
pbar.close()
try:
db.close()
except AttributeError:
pass
return db, nframes
def analyze(imgpath, model):
imgname = os.path.basename(imgpath)
kernelSize = 64
kernelStepSize = 1
bufferVal = 8 # will load kernelSize x bufferVal
stepsize = 1
starttime = time.time()
slide = openslide.open_slide(imgpath)
#tissue detection
thumbnail = np.array(
slide.get_thumbnail((slide.level_dimensions[0][0] / 64,
slide.level_dimensions[0][1] / 64)))
thumbnailGray = color.rgb2gray(thumbnail)
val = filters.threshold_otsu(thumbnailGray)
tissueMask = thumbnailGray < max(val, 0.8)
plt.imsave('tissue.png', tissueMask) #save the thumb of tissue mask
buffersize = kernelSize * bufferVal
resultMask = thumbnail.astype(numpy.uint8) * 0
if stepsize > 1:
resultMask = np.resize(
resultMask,
(int(resultMask.shape[0] / stepsize),
int(resultMask.shape[1] / stepsize), resultMask.shape[2]))
counter1 = 0
counter2 = 0
expectedStep = tissueMask.shape[0] / bufferVal
outputsVec = []
for i in range(0, tissueMask.shape[0] - bufferVal, bufferVal): #Height
curMod = i % (bufferVal * max(5, int(expectedStep / 20)))
if curMod == 0:
print('.', end='', flush=True)
for j in range(0, tissueMask.shape[1] - bufferVal, bufferVal): #Width
if np.mean(tissueMask[i:i + bufferVal, j:j + bufferVal]) < (
8 / 16): #most of them are background
continue
bigTile = numpy.array(
slide.read_region((j * kernelSize, i * kernelSize), 0,
[buffersize, buffersize]))
bigTile = color.rgba2rgb(bigTile)
sz = bigTile.itemsize
h, w, cs = bigTile.shape
bh, bw = kernelSize, kernelSize
shape = (int(h / bh / stepsize), int(w / bw / stepsize), bh, bw,
cs)
strides = (stepsize * w * bh * sz * cs, stepsize * sz * cs * bw,
w * sz * cs, sz * cs, sz)
blocks = np.lib.stride_tricks.as_strided(bigTile,
shape=shape,
strides=strides)
blocks = blocks.reshape(blocks.shape[0] * blocks.shape[1],
blocks.shape[2], blocks.shape[3],
blocks.shape[4])
predictions = model.predict(blocks)
outputsVec = outputsVec + predictions
qwe = np.array(predictions)
qwe = qwe.reshape(int(h / bh / stepsize), int(w / bw / stepsize),
2)
counter1 = counter1 + sum(np.array(predictions)[:, 1] > 0.5)
counter2 = counter2 + len(predictions) - sum(
np.array(predictions)[:, 1] > 0.5)
resultMask[int(i / stepsize):int((i + bufferVal) / stepsize),
int(j / stepsize):int((j + bufferVal) / stepsize),
0] = 255 * qwe[:, :, 1]
resultMask[int(i / stepsize):int((bufferVal + i) / stepsize),
int(j / stepsize):int((bufferVal + j) / stepsize),
1] = 255 * qwe[:, :, 0]
endtime = time.time()
elapsedtime = endtime - starttime
print('elapsed time ' + str(elapsedtime))
outputname = 'output/' + imgname + '-f' + str(counter2) + '-o' + str(
counter1) + '.png'
plt.imsave(outputname, resultMask, cmap=plt.cm.gray)
outputname2 = 'output/' + imgname + '-f' + str(counter2) + '-o' + str(
counter1) + '.csv'
writeXML(outputname2, outputsVec)
return (counter2, counter1)
def recommend_similarity_item(item):
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
### hyper parameter
k = 3 ### 출력할 유사한 item의 갯수
rows = 1 ### 출력할 item image의 행
cols = k ### 출력할 item image의 열
space = 1 ### 출력할 item이 놓일 위치
### 경로 설정
root_path = "../../experiment/" ### file 경로의 공통된 root_path
feature_path = "data/polyvore/dataset/" ### feature vector를 저장한 파일의 위치
train_json = 'data/polyvore/jsons/train_no_dup.json' ### train data의 위치
valid_json = 'data/polyvore/jsons/valid_no_dup.json' ### validation data의 위치
test_json = 'data/polyvore/jsons/test_no_dup.json' ### test data의 위치
image_folder_path = 'data/polyvore/images/' ### dataset의 image들이 저장된 위치
### item 간의 유사도를 계산하기 위한 cosine similarity 함수
def cos_sim(A, B):
return np.dot(A,B)/(np.linalg.norm(A)*np.linalg.norm(B))
### 미리 학습한 training set에 있는 item들의 feature vector 불러오기
with open(root_path+feature_path+'imgs_featdict_train.pkl', 'rb') as train_feat:
train_data = pickle.load(train_feat)
### 미리 학습한 validation set에 있는 item들의 feature vector 불러오기
with open(root_path+feature_path+'imgs_featdict_valid.pkl', 'rb') as valid_feat:
valid_data = pickle.load(valid_feat)
### 미리 학습한 test set에 있는 item들의 feature vector 불러오기
with open(root_path+feature_path+'imgs_featdict_test.pkl', 'rb') as test_feat:
test_data = pickle.load(test_feat)
save_feat = [] ### save_id: 각 item을 구별할 수 있는 id만 저장하기 위한 list
save_id = [] ### save_feat: 각 item의 feature vector만 저장하기 위한 list
### train data에 대하여...
for key, value in train_data.items():
save_feat.append(value)
save_id.append(key)
### validation data에 대하여...
for key, value in valid_data.items():
save_feat.append(value)
save_id.append(key)
### test data에 대하여...
for key, value in test_data.items():
save_feat.append(value)
save_id.append(key)
print('nums of total item: {}\n'.format(len(save_id)))
###############################################
### 모든 item들의 id와 feature vector를 저장 ###
###############################################
data_dict = {} ### data_dict은 전체 data에 있는 item에 대하여
### key: item_id, value: feature vector를 갖는다.
for i in range(len(save_id)):
data_dict[save_id[i]] = save_feat[i]
###############################################
###############################################################################
### user가 등록한 이미지 출력 ###
###############################################################################
### 임의의 user item 등록
item_path = item
### 등록 이미지 출력
user_img = cv2.imread(item_path)
user_img = cv2.resize(user_img, dsize=(256,256), interpolation=cv2.INTER_AREA)
user_img = plt.imshow(cv2.cvtColor(user_img, cv2.COLOR_BGR2RGB))
plt.axis("off")
###############################################################################
print("Start loading pretrained resnet50 model...")
model = models.resnet50(pretrained=True)
model = nn.Sequential(*list(model.children())[:-1])
model = model.to(device)
model.eval()
print("Finish loading pretrained restnet50 model!!!\n")
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
transform = transforms.Compose([
transforms.ToPILImage(),
transforms.Resize(256), transforms.CenterCrop(224),
transforms.ToTensor(), normalize
])
### resnet50을 통해 user image의 feature vector를 추출하는 함수
def process_image(im):
im = transform(im)
im = im.unsqueeze_(0)
im = im.to(device)
out = model(im)
return out.squeeze()
print("Start extract user items feature vector")
with torch.no_grad():
im = skimage.io.imread(item_path)
if len(im.shape) == 2:
im = gray2rgb(im)
if im.shape[2] == 4:
im = rgba2rgb(im)
im = resize(im, (256,256))
im = img_as_ubyte(im)
feats = process_image(im).cpu().numpy()
print("Finish extract user items feature vector\n")
res_list = [] ### res_list에는 user의 item과 data의 item의 유사도를 저장하는 list
res_dict = {} ### key: polyvore item, value: similarity score
print("Start calculating similarity scores...")
for i in range(len(save_feat)):
res = cos_sim(feats,save_feat[i])
res_list.append(res)
res_dict[save_id[i]] = res_list[i]
print("Finish calculating simiarity scores!!!\n")
print("Start sorting similarity score...")
sort_list = sorted(res_dict.items(), key=lambda item:item[1], reverse=True)
print("Finish sorting similarity score!!!\n")
best_sim_list = [] ### k개의 가장 유사성이 높은 item을 저장하는 list
for i in range (k):
best_sim_list.append(sort_list[i][0])
######################################################
### json data load ###
######################################################
### train.json에 대하여...
with open(root_path+train_json, "r") as train_file:
train_in_outfit_id = json.load(train_file)
### valid.json에 대하여...
with open(root_path+valid_json, "r") as valid_file:
valid_in_outfit_id = json.load(valid_file)
### test.json에 대하여...
with open(root_path+test_json, "r") as test_file:
test_in_outfit_id = json.load(test_file)
#######################################################
outfit2item = {} ### outfit2item은 outfit을 조합하는 item을 저장한다.
### key: outfit_id, value: outfit을 구성하는 item id들
outfit2index = {} ### outf2index는 outfit을 조합하는 index를 저장한다.
### key: outfit_id, value: outfit을 구성하는 index들
################################################################################
### 각 outfit 마다 item_id와 index를 딕셔너리 형태로 각각 저장 ###
################################################################################
### 17,316개의 outfit을 갖고 있는 train set에ㅔ 대하여...
for i in range(len(train_in_outfit_id)): ### len(train_in_outfit_id: 17316)
outfit_id = train_in_outfit_id[i]["set_id"]
item_index = [] ### train data에서 outfit을 구성하는 index를 저장하는 list
item_id = [] ### train data에서 outfit을 구성하는 item id를 저장하는 list
for j in range(len(train_in_outfit_id[i]["items"])):
index = train_in_outfit_id[i]["items"][j]["index"]
item_index.append(index)
_, each_item_id = train_in_outfit_id[i]["items"][j]["image"].split('id=')
item_id.append(each_item_id)
outfit2index[outfit_id] = item_index
outfit2item[outfit_id] = item_id
### 1,497개의 outfit을 갖고 있는 validation set에 대하여...
for i in range(len(valid_in_outfit_id)): ### len(train_in_outfit_id: 17316)
outfit_id = valid_in_outfit_id[i]["set_id"]
item_index = [] ### validation data에서 outfit을 구성하는 index를 저장하는 list
item_id = [] ### validation data에서 outfit을 구성하는 item id를 저장하는 list
for j in range(len(valid_in_outfit_id[i]["items"])):
index = valid_in_outfit_id[i]["items"][j]["index"]
item_index.append(index)
_, each_item_id = valid_in_outfit_id[i]["items"][j]["image"].split('id=')
item_id.append(each_item_id)
outfit2index[outfit_id] = item_index
outfit2item[outfit_id] = item_id
### 3,076개의 outfit을 갖고 있는 test set에 대하여....
for i in range(len(test_in_outfit_id)): ### len(train_in_outfit_id: 17316)
outfit_id = test_in_outfit_id[i]["set_id"]
item_index = [] ### test data에서 outfit을 구성하는 index를 저장하는 list
item_id = [] ### test data에서 outfit을 구성하는 item id를 저장하는 list
for j in range(len(test_in_outfit_id[i]["items"])):
index = test_in_outfit_id[i]["items"][j]["index"]
item_index.append(index)
_, each_item_id = test_in_outfit_id[i]["items"][j]["image"].split('id=')
item_id.append(each_item_id)
outfit2index[outfit_id] = item_index
outfit2item[outfit_id] = item_id
################################################################################
outfit_id = [] ### outfit의 id를 저장하는 list
index_id = [] ### outfit의 index 갯수를 저장하는 list
for i in range(k):
for key in outfit2item.keys():
for j in range(len(outfit2item[key])):
if best_sim_list[i] == outfit2item[key][j]:
outfit_id.append(str(key))
index_id.append(str(j+1))
sim_save_file = [] ### 유사한 item의 image 파일 경로를 저장하는 list
for i in range(k):
sim_save_file.append(image_folder_path+outfit_id[i]+'/'+index_id[i]+'.jpg')
### http://blog.daum.net/geoscience/1263 참고
fig = plt.figure()
for i in range(k):
img = cv2.imread(root_path+sim_save_file[i])
img = cv2.resize(img, dsize=(256,256), interpolation=cv2.INTER_AREA)
ax = fig.add_subplot(rows,cols,space)
ax.imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
space += 1
ax.axis("off")
plt.show()
from skimage import io
imagergba = io.imread("/Users/sameriksson/temp/screensgen/1.png")
from skimage.color import rgba2rgb
image = rgba2rgb(imagergba)
from skimage import transform
img = transform.resize(image, (224, 224)) #224
import matplotlib.pyplot as plt
plt.imshow(img)
plt.show()
numpy_image = img.transpose((2, 0, 1))
print(numpy_image.shape)
import torch.nn as nn
conv1 = nn.Conv2d(3, 6, 5)
print(conv1)
#Conv2d(3, 6, kernel_size=(16, 16), stride=(1, 1))
import torch
tensor_image = torch.from_numpy(numpy_image)
tensor_images = tensor_image.unsqueeze(0)
conv1 = conv1.double()
c = conv1(tensor_images)
print(c.shape)
import torch.nn.functional as F
b = F.relu(c)
pool = nn.MaxPool2d(2, 2)
d = pool(b)
print(d.shape)
conv2 = nn.Conv2d(6, 16, 5)
conv2 = conv2.double()
e = conv2(d)
print(e.shape)
f = F.relu(e)
D = np.loadtxt('dltrainfiles/dl4_rgb_ds16_maps.txt')
# D = np.loadtxt('dl8_rgb_ds192.txt')
# D = io.imread('dict.png').astype(float) / 65535
# dl.visualize_dictionary(D, 16, 8)
dir_images = "/home/eduardo/Imagens/*.png"
images = glob.glob(dir_images)
for i, img in enumerate(images):
print(i, img)
imgidx = int(input("Escolha do id da imagem: "))
# img_train = io.imread(images[imgidx], as_grey=True)
image = io.imread(images[imgidx])
if image.shape[-1] == 4:
image = color.rgba2rgb(image)
if image.max() > 1.0:
img_train = image / 255.
else:
img_train = image
# img_train = io.imread(images[imgidx])[:,:,:3]
nl, nc, _ = img_train.shape
ml = nl % m11
mc = nc % m22
img_train = img_train[:(nl - ml), :(nc - mc), :].astype(float)
# print(color.rgba2rgb(io.imread(images[imgidx]))[0,0])
# io.imshow(color.yuv2rgb(img_train))
# io.show()
def predict(model: nn.Module, slide_reader: OpenSlide, img_id, bbox,
scale_factor):
"""
:param model:
:param slide_reader:
:param img_id:
:param bbox:
:param scale_factor:
:return:
"""
minr, minc, maxr, maxc = bbox
scf_row, scf_col = scale_factor
row_st_origin, col_st_origin = int(minr * scf_row), int(minc * scf_col)
row_ed_origin, col_ed_origin = int(maxr * scf_row), int(maxc * scf_col)
row_length, col_length = row_ed_origin - row_st_origin, col_ed_origin - col_st_origin
if row_length < 500 or col_length < 500:
return
# print('read region')
# print(row_st_origin, col_st_origin)
# print((col_length + (alpha - 1) * PATCH_STRIDE, row_length + (alpha - 1) * PATCH_STRIDE))
if (row_length - PATCH_SIZE - (alpha - 1) * SD) % PATCH_STRIDE != 0:
row_ed_origin += PATCH_STRIDE - (row_length - PATCH_SIZE -
(alpha - 1) * SD) % PATCH_STRIDE
if (col_length - PATCH_SIZE - (alpha - 1) * SD) % PATCH_STRIDE != 0:
col_ed_origin += PATCH_STRIDE - (col_length - PATCH_SIZE -
(alpha - 1) * SD) % PATCH_STRIDE
# dpt_table
# whole_dpt = []
# cm_hot = mpl.cm.get_cmap('bwr')
cm_hot = mpl.cm.get_cmap('coolwarm')
patch_loader = data_utils.DataLoader(dataset=PredictPatchLoader(
slide_reader, row_st_origin, row_ed_origin, col_st_origin,
col_ed_origin, PATCH_SIZE, PATCH_STRIDE, alpha, SD, img_id),
batch_size=1,
num_workers=3)
cols = (col_ed_origin - col_st_origin - PATCH_SIZE -
(alpha - 1) * SD) // PATCH_STRIDE + 1
all_dpt = []
dpt_table_row = []
t = time_sp()
for data in patch_loader:
# t.time_span('转为Tensor')
# row_idx = idx // cols
# col_idx = idx % cols
data.squeeze_()
opt_table = []
with torch.no_grad():
data = data.float()
for opt_row in range(alpha):
opt_table.append([])
for opt_col in range(alpha):
current_data = data[opt_row * alpha + opt_col, :, :, :]
current_data.unsqueeze_(0)
# print(current_data.shape)
# t.time_span('Tensor 切片')
outputs = model(current_data)
opts = nn.functional.softmax(outputs, dim=1).cpu().numpy()
opts = opts[0, 1, :, :]
# print(opts.shape)
opt_table[-1].append(opts)
# t.time_span('预测耗时')
dpt_table = merge_opt(opt_table)
# t.time_span('Merge OPT')
dpt_table_row.append(dpt_table)
if len(dpt_table_row) == cols:
row_img = np.concatenate(dpt_table_row, axis=1)
all_dpt.append(row_img)
# for each in all_dpt:
# print(each.shape)
# print(f'all_apt len {len(all_dpt)}')
dpt_table_row = []
if len(all_dpt) == 0:
print(bbox)
return
whole_pred = np.concatenate(all_dpt)
# with open('dpt.dump', 'wb') as outfile:
# pickle.dump(whole_pred, outfile)
htop_map = cm_hot(whole_pred)
dims = slide_reader.level_dimensions
down_level = -5
fr = dims[0][1] / dims[down_level][1]
fc = dims[0][0] / dims[down_level][0]
# d = int( PATCH_SIZE + (alpha-1) * PATCH_STRIDE / fc)
# print(
# (col_st_origin , row_st_origin),
# len(dims) - 5,
# (int((col_ed_origin - col_st_origin) // fc), int((row_ed_origin - row_st_origin) // fr))
# )
region_img = slide_reader.read_region(
(col_st_origin, row_st_origin),
len(dims) + down_level,
(int((col_ed_origin - col_st_origin) // fc),
int((row_ed_origin - row_st_origin) // fr))).convert('RGB')
# region_img = thumbnail[minr:maxr, minc:maxc]
tr, tc = whole_pred.shape
region_img = np.asarray(region_img)
region_img = transform.resize(region_img, (tr, tc))
# region_img = np.resize(region_img, (tr, tc, 3))
save_image = 0.7 * region_img + 0.3 * color.rgba2rgb(htop_map)
save_image = save_image.clip(0, 1.0)
info = '_'.join(
map(str, (row_st_origin, col_st_origin, row_ed_origin - row_st_origin,
col_ed_origin - col_st_origin)))
if not path.exists(f'pred/{img_id}'):
os.mkdir(f'pred/{img_id}')
np.save(f'pred/{img_id}/{info}.npy', whole_pred)
io.imsave(f'pred/{img_id}/{info}_gray.bmp', whole_pred)
io.imsave(f'pred/{img_id}/{info}_region.bmp', region_img)
io.imsave(f'pred/{img_id}/{info}_heatmap.bmp', htop_map)
io.imsave(f'pred/{img_id}/{info}.bmp', save_image)
def load_img(self, idx):
img_rgb_path = self.rgb_paths[idx]
img = io.imread(img_rgb_path)
img = color.rgba2rgb(img)
# img = img / 255.
return img
def determine_skew_dev(
image: ImageType,
sigma: float = 3.0,
num_peaks: int = 20,
num_angles: int = 180
) -> Tuple[Optional[np.float64], List[List[np.float64]], np.float64, Tuple[
ImageTypeUint64, List[List[np.float64]], ImageTypeFloat64], ]:
"""Calculate skew angle."""
imagergb = rgba2rgb(image) if len(
image.shape) == 3 and image.shape[2] == 4 else image
img = rgb2gray(imagergb) if len(imagergb.shape) == 3 else imagergb
edges = canny(img, sigma=sigma)
out, angles, distances = hough_line(
edges, np.linspace(-np.pi / 2, np.pi / 2, num_angles, endpoint=False))
hough_line_out = (out, angles, distances)
_, angles_peaks, _ = hough_line_peaks(out,
angles,
distances,
num_peaks=num_peaks,
threshold=0.05 * np.max(out))
absolute_deviations = [_calculate_deviation(k) for k in angles_peaks]
average_deviation: np.float64 = np.mean(np.rad2deg(absolute_deviations))
angles_peaks_degree = [np.rad2deg(x) for x in angles_peaks]
bin_0_45 = []
bin_45_90 = []
bin_0_45n = []
bin_45_90n = []
for angle in angles_peaks_degree:
deviation_sum = int(90 - angle + average_deviation)
if _compare_sum(deviation_sum):
bin_45_90.append(angle)
continue
deviation_sum = int(angle + average_deviation)
if _compare_sum(deviation_sum):
bin_0_45.append(angle)
continue
deviation_sum = int(-angle + average_deviation)
if _compare_sum(deviation_sum):
bin_0_45n.append(angle)
continue
deviation_sum = int(90 + angle + average_deviation)
if _compare_sum(deviation_sum):
bin_45_90n.append(angle)
angles = [bin_0_45, bin_45_90, bin_0_45n, bin_45_90n]
nb_angles_max = 0
max_angle_index = -1
for angle_index, angle in enumerate(angles):
nb_angles = len(angle)
if nb_angles > nb_angles_max:
nb_angles_max = nb_angles
max_angle_index = angle_index
if nb_angles_max:
ans_arr = _get_max_freq_elem(angles[max_angle_index])
angle = np.mean(ans_arr)
elif angles_peaks_degree:
ans_arr = _get_max_freq_elem(angles_peaks_degree)
angle = np.mean(ans_arr)
else:
return None, angles, average_deviation, hough_line_out
rot_angle = (angle + 45) % 90 - 45
return rot_angle, angles, average_deviation, hough_line_out
def extract_trajectories(
image_filenames, point_method_kwargs={}, flow_method_kwargs={}
):
"""
Times where no trajectory point are found will have index -1
"""
traj_points = []
traj_files = []
img_prev_arr = None
for fn in image_filenames:
img_src = Image.open(fn)
img_gray_arr = np.array(rgb2gray(rgba2rgb(np.array(img_src))))
if img_prev_arr is None:
points = shitomasi_detection(img_gray_arr, **point_method_kwargs)
if len(points) < MIN_CORNERS:
continue
else:
traj_points.append(points)
traj_files.append(fn)
else:
points = traj_points[-1]
xy_end = track_features(
img_prev_arr, img_gray_arr, points=points, **flow_method_kwargs
)
traj_points.append(xy_end)
traj_files.append(fn)
img_prev_arr = img_gray_arr
traj_points = np.array(traj_points)
# NB: images use (x,y) indexing order rather than (i,j)
ny_img, nx_img = img_gray_arr.shape
# make all invalid points have index -1 and round to nearest integer
np.nan_to_num(traj_points, copy=False, nan=-1.0)
traj_points = traj_points.round().astype(int)
# y-indexing is from top-left for images
traj_points[..., 1] = ny_img - 1 - traj_points[..., 1]
# now we need to clean up the points. Some will be outside the valid range
# [0, Nx] and [0, Ny]
m_invalid = np.logical_or(
# out of x index bounds
np.logical_or(traj_points[..., 0] < 0, traj_points[..., 0] >= nx_img),
# out of y index bounds
np.logical_or(traj_points[..., 1] < 0, traj_points[..., 1] >= ny_img),
)
# make these all -1 (indicating invalid point)
traj_points[..., 0] = np.where(m_invalid, -1, traj_points[..., 0])
traj_points[..., 1] = np.where(m_invalid, -1, traj_points[..., 1])
N_img, N_trajs, _ = traj_points.shape
assert len(traj_files) == N_img
ds = xr.Dataset(coords=dict(image_filename=traj_files, traj_id=np.arange(N_trajs)))
ds["i"] = ("image_filename", "traj_id"), traj_points[..., 0]
ds["i"].attrs["long_name"] = "x-index"
ds["i"].attrs["units"] = "1"
ds["j"] = ("image_filename", "traj_id"), traj_points[..., 1]
ds["j"].attrs["long_name"] = "j-index"
ds["j"].attrs["units"] = "1"
return ds
edges = canny(img_hsv_hue)
# fill holes
fill_holes = ndi.binary_fill_holes(edges)
# label, 填充
structure = [[1, 1, 1], [1, 1, 1], [1, 1, 1]]
label_objects, nb_labels = ndi.label(fill_holes, structure)
# 剔除额外的标签
sizes = np.bincount(label_objects.ravel())
mask_sizes = sizes > (height * width) * 1e-5
mask_sizes[0] = 0
bin_img_cleaned = mask_sizes[label_objects]
# 提取边界点
return extract_edges_points(bin_img_cleaned)
if __name__ == "__main__":
from skimage import io
img = io.imread('./img1.png')
channel = img.shape[2]
if channel == 4:
img_rgb = rgba2rgb(img)
points, edges_mask = get_edges_from_colouration(img_rgb)
# 结果呈现
# plt.plot(points[0],points[1],'.') # 根据坐标点画边界曲线
show_with_mask(img_rgb, edges_mask)
filename = f"{item['id']}.jpg"
image_path = os.path.join(args.image_dir, filename)
if os.path.exists(image_path):
try:
img = skimage.io.imread(image_path)
except Exception as e:
print(e)
continue
else:
continue
img = np.array(img)
if len(img.shape) == 2:
img = gray2rgb(img)
if img.shape[2] == 4:
img = rgba2rgb(img)
img = resize(img, (256, 256))
img = img_as_ubyte(img)
feats = process_image(img).cpu().numpy()
features[item['id']] = feats
print(f'Total: {len(features)}')
with open(save_dict, 'wb') as handle:
pkl.dump(features, handle, protocol=pkl.HIGHEST_PROTOCOL)
# filter images
if 0 < n_images < len(data_paths):
data_paths = data_paths[:n_images]
# repeat data
elif n_images > len(data_paths):
repeats = n_images // len(data_paths) + 1
data_paths = base_data_paths * repeats
data_paths = data_paths[:n_images]
with Timer(store=stores.get("data pre-processing", []),
ignore=time_inference):
# read images to numpy arrays
data = [io.imread(str(d)) for d in data_paths]
# rgba to rgb
data = [
im if im.shape[-1] == 3 else uint8(rgba2rgb(im) * 255)
for im in data
]
# resize images to target_size or
if keep_ratio:
# to closest multiples of 128 <= max_im_width, keeping aspect ratio
new_sizes = [to_128(d, max_im_width) for d in data]
data = [
resize(d, ns, anti_aliasing=True)
for d, ns in zip(data, new_sizes)
]
else:
# to args.target_size
data = [resize_and_crop(d, target_size) for d in data]
new_sizes = [(target_size, target_size) for _ in data]
# resize() produces [0, 1] images, rescale to [-1, 1]
