def job(query: str):
message = None
if not query.startswith('test'):
job_id = utils.get_job_from_string(query)
else:
job_id = query
job = utils.get_job(job_id)
if not job:
message = f'There isn`t job for word: "{query}"'
else:
if job.get_error_path():
message = f'Error for job for word: "{query}"'
if message:
flash(message, 'errors')
return redirect(url_for('main'))
try:
job_results = utils.apply_window(utils.get_results(job))
except FileNotFoundError:
return render_template("job.html",
meta=job.get_meta(),
inprogress=True,
data=[],
labels=[])
return render_template("job.html",
meta=job.get_meta(),
data=job_results.data,
labels=list(map(str, job_results.labels)))
def get_test_data(batch_size):
file_dir = root_dir + 'severance_data/imgs/'
input_list_pkl = root_dir + 'severance_data/pixel_diff/val_x.pkl'
target_list_pkl = root_dir + 'severance_data/pixel_diff/val_y.pkl'
with open(input_list_pkl, 'rb') as f:
full_input_list = pickle.load(f)
with open(target_list_pkl, 'rb') as f:
full_target_list = pickle.load(f)
assert (len(full_input_list) == len(full_target_list))
indexes = np.arange(len(full_input_list))
np.random.shuffle(indexes)
full_input_list = np.array(full_input_list)[indexes]
full_target_list = np.array(full_target_list)[indexes]
input_imgs = []
target_imgs = []
for x, y in zip(full_input_list[:batch_size],
full_target_list[:batch_size]):
input_imgs.append(apply_window(np.load(file_dir + x + '.npy')))
target_imgs.append(apply_window(np.load(file_dir + y + '.npy')))
return np.array(input_imgs), np.array(target_imgs)
def get_train_pair(batch_size):
file_dir = root_dir + 'severance_data/supervision/'
d_list = os.listdir(file_dir)
xs = []
ys = []
np.random.shuffle(d_list)
for dname in d_list[:batch_size]:
ys.append(np.load(file_dir + dname))
img_name = root_dir + 'severance_data/imgs/' + dname
img = np.load(img_name)
img = apply_window(img)
xs.append(img)
return np.array(xs), np.array(ys)
for x, y in zip(full_input_list, full_target_list):
d1 = pydicom.dcmread(file_dir + x)
d2 = pydicom.dcmread(file_dir + y)
input_imgs.append(d1.pixel_array)
target_imgs.append(d2.pixel_array)
if wo_name:
return np.array(input_imgs), np.array(target_imgs)
else:
return np.array(input_imgs), np.array(target_imgs), full_input_list
batch_size = 10
x, y, names = get_data_all(wo_name=False)
supervisions = []
for i in range(x.shape[0]):
start = time.time()
a = apply_window(x[i])
b = apply_window(y[i])
mask = get_mask(a)
mask2 = get_local_(mask, 'max', 5)
mask2 = get_local_(mask2, 'min', 7)
diff = pixel_diff(a, b)
# supervisions.append(mask2 * diff)
end = time.time()
print('time per img: %.3f, step: %5d' % (end - start, i))
np.save('supervision/' + names[i], mask2 * diff)