def show_siamese_batch(loader):
img_batch1, img_batch2, label_batch = next(iter(loader))
print("Image batch size: {}. Label batch size: {}.".format(
img_batch1.size(), len(label_batch)))
print("Average pixel value in batch: {:.5f}".format(
(img_batch1.mean() + img_batch1.mean()) / 2.))
print("Stddev pixel value in batch: {:.5f}".format(
(img_batch1.std() + img_batch2.std()) / 2.))
# display batch
n_cols = 8
n_rows = 2 * (img_batch1.size()[0] // n_cols + 1 *
(img_batch1.size()[0] % n_cols > 0))
fig, axs = plt.subplots(n_rows, n_cols, figsize=(15, n_rows * 15 / n_cols))
for i in range(n_rows // 2):
for j in range(n_cols):
k = i * n_cols + j
if k < img_batch1.size()[0]:
img1 = img_batch1[k].permute(1, 2, 0).data.cpu().numpy()
img2 = img_batch2[k].permute(1, 2, 0).data.cpu().numpy()
img1, img2 = imgMinMaxScaler(img1), imgMinMaxScaler(img2)
label = label_batch[k]
axs[2 * i, j].imshow(rio.convert_tensor_to_rgb(img1))
axs[2 * i + 1, j].imshow(rio.convert_tensor_to_rgb(img2))
axs[2 * i, j].set_title(label)
axs[2 * i + 1, j].set_title(label)
plt.show()
def __getitem__(self, index):
item = self.items[index].copy()
bunch = item.pop('images')
if self.channels_mode == 'six':
channels = []
for i, filename in sorted(bunch, key=itemgetter(0)):
img = self.open_fn(filename)
img = img if self.tr is None else self.tr(img)
channels.append(img)
sample = torch.cat(channels, dim=0)
else:
filenames = [filename for _, filename in bunch]
t = rio.load_images_as_tensor(filenames)
arr = rio.convert_tensor_to_rgb(t)
img = PIL.Image.fromarray(arr.astype(np.uint8))
sample = img if self.tr is None else self.tr(img)
# y = self.targets[index]
y = item[self.targets_key]
if self.drop_meta:
item = dict(features=sample, targets=y)
else:
item.update(dict(features=sample, targets=y))
if self.onehot:
y_enc = get_one_hot(y,
smoothing=self.label_smoothing,
num_classes=self.num_classes)
sample['targets_one_hot'] = y_enc
return item
def _show_imgs(self, imgs):
from matplotlib import pyplot as plt
import rxrx.io as rio
import cv2
for i, img in enumerate(imgs):
img = np.moveaxis(img, 0, 2)
height, width, _ = img.shape
img = cv2.resize(img,
dsize=(512, 512),
interpolation=cv2.INTER_CUBIC)
img_rgb = np.array(rio.convert_tensor_to_rgb(img), dtype='uint8')
img_rgb = cv2.resize(img_rgb,
dsize=(height, width),
interpolation=cv2.INTER_CUBIC)
ax = plt.subplot(1, len(imgs), i + 1)
ax.title.set_text(self.mode)
plt.imshow(img_rgb)
plt.show()
def plot_rgb(data, figsize=10):
"""
convert 6-channels to RGB (3-channel)
and visualize the figure.
"""
dirs = data[0].split("/")
base_plates = [dirs.index(i) for i in dirs if i.startswith("Plate")][0]
pre = "_".join(dirs[base_plates - 1:base_plates + 1])
suf = "_".join(os.path.basename(data[0]).split("_")[:2])
imgname = "_".join((pre, suf))
plt.figure(figsize=(figsize, figsize))
img_6channel = rio.load_images_as_tensor(data)
img_as_rgb = rio.convert_tensor_to_rgb(img_6channel)
plt.title(imgname)
plt.imshow(img_as_rgb)
def plot_cell_sirna_rgb(filtered_data, figsize=20):
"""
plot all data in a row (Actually mutliple rows)
:param filtered_data: dataframe, after processed with 'filter_data_with_cell_sirna'
:param figsize: column size
:return:
"""
data_num = len(filtered_data)
row_num = math.ceil((data_num // 4))
plt.figure(figsize=(figsize, figsize * row_num * 0.25))
for i in range(data_num):
plt.subplot(row_num, 4, i + 1)
img_paths = filtered_data.iloc[i, -6:].tolist()
img_6channel = rio.load_images_as_tensor(img_paths)
img_as_rgb = rio.convert_tensor_to_rgb(img_6channel)
plt.title(filtered_data.iat[i, 0])
plt.imshow(img_as_rgb)
base_path=LOCAL_IMAGES_BASE_PATH)
t.shape
# In[5]:
fig, axes = plt.subplots(2, 3, figsize=(24, 16))
for i, ax in enumerate(axes.flatten()):
ax.axis('off')
ax.set_title('channel {}'.format(i + 1))
_ = ax.imshow(t[:, :, i], cmap='gray')
# In[6]:
x = rio.convert_tensor_to_rgb(t)
x.shape
# In[7]:
plt.figure(figsize=(8, 8))
plt.axis('off')
_ = plt.imshow(x)
# In[8]:
y = rio.load_site_as_rgb('train',
'HUVEC-08',
4,
def upload():
print('--> /upload')
if 'POST' == request.method:
HEPG2 = request.form.get('radio-choice-h-2a')
HVUEC = request.form.get('radio-choice-h-2b')
RPE = request.form.get('radio-choice-h-2c')
U2OS = request.form.get('radio-choice-h-2d')
cell_type = ''
if HEPG2:
cell_type = 'HEPG2'
elif HVUEC:
cell_type = 'HVUEC'
elif RPE:
cell_type = 'RPE'
else:
cell_type = 'U2OS'
experiment_on = request.form.get('experiment-no')
plate_no = request.form.get('plate-no')
positive_control = request.form.get('well-radio-choice-h-2')
negative_control = request.form.get('well-radio-choice-h-2b')
treatment = request.form.get('well-radio-choice-h-2c')
well_type = ''
if positive_control:
well_type = 'positive_control'
elif negative_control:
well_type = 'negative_control'
else:
well_type = 'treatment'
print(HEPG2, HVUEC, RPE, U2OS)
print(experiment_on)
files = []
rgb_file_name = ''
for file in request.files.getlist('files'):
if file and allowed_file(file.filename):
filename = cell_type + '-' + experiment_on \
+ '-Plate' + plate_no \
+ '-' + secure_filename(file.filename)
print('-->', file.filename, '==>', filename)
t = filename.split('_')
rgb_file_name = t[0] + '_rgb.png'
files.append(filename)
file.save(os.path.join(app.config['UPLOAD_FOLDER'], filename))
# load
channel_paths = [
os.path.join(app.config['UPLOAD_FOLDER'], f) for f in files
]
tensor = rio.load_images_as_tensor(channel_paths)
#
rgb = rio.convert_tensor_to_rgb(tensor)
rgb = np.array(rgb, dtype=np.uint8)
matplotlib.image.imsave(
os.path.join(app.config['UPLOAD_FOLDER'], rgb_file_name), rgb)
# predict:
tensor = np.expand_dims(tensor, axis=0)
predictions = pred({'feature': tensor})
print(predictions)
id = str(uuid.uuid1())
submit_time = isoformat(datetime.datetime.now())
# save file to database
record_file = File(id, session['email'], cell_type, experiment_on,
plate_no, well_type, ','.join(files), rgb_file_name,
submit_time, '%d' % predictions['classes'][0])
db.session.add(record_file)
db.session.commit()
return redirect('file_list.html')