def encode_labels(labels: List[List[str]], remove_too_long: bool = True) -> List[List[int]]:
encoded_labels = list(map(lambda t: [0] + list(map(lambda tt: CLASSES.index(tt), t)) + [0],
labels))
if remove_too_long:
encoded_labels = list(filter(lambda i: len(i) <= MAX_LEN, encoded_labels))
encoded_labels = tf.keras.preprocessing.sequence.pad_sequences(encoded_labels,
maxlen=MAX_LEN,
padding='post',
truncating='post',
value=0)
encoded_labels = tf.keras.utils.to_categorical(encoded_labels, len(CLASSES))
return encoded_labels
def read_kitti_label(fpath):
objs = []
with open(fpath, 'r') as fin:
for line in fin:
cols = line.split(' ')
if cols[0] == 'DontCare':
continue
objs.append({
'class':
CLASSES.index(cols[0]),
'bbox': [
float(cols[4]),
float(cols[5]),
float(cols[6]),
float(cols[7])
] # left, top, right, bottom
})
return objs
def get_legend_color(class_name : int):
"""
Возвращает цвет ячейки для `pandas.Styler` при создании легенды.
Раскарасит ячейку те же цветом, который имеют боксы соотвествующего класс
Arguments
---------
class_name : int
название класса согласно списку классов MS COCO
Returns
-------
str
background-color для ячейки, содержащей class_name
"""
index = CLASSES.index(class_name)
color = rgb_colors[index]
return 'background-color: rgb({color[0]},{color[1]},{color[2]})'.format(color=color)
def read_data(xml_path, img_path):
tree = ET.parse(xml_path)
root = tree.getroot()
objects = root.findall("object")
names = []
gtbboxes = np.zeros([len(objects), 4], dtype=np.int32)
for idx, obj in enumerate(objects):
names.append(obj.find("name").text)
xmin = int(obj.find("bndbox").find("xmin").text)
xmax = int(obj.find("bndbox").find("xmax").text)
ymin = int(obj.find("bndbox").find("ymin").text)
ymax = int(obj.find("bndbox").find("ymax").text)
gtbboxes[idx, 0] = (xmin + xmax) // 2
gtbboxes[idx, 1] = (ymin + ymax) // 2
gtbboxes[idx, 2] = xmax - xmin
gtbboxes[idx, 3] = ymax - ymin
img = np.array(Image.open(img_path))
labels = np.zeros([len(objects)])
for idx, name in enumerate(names):
labels[idx] = CLASSES.index(name)
return img, gtbboxes, labels
def predict(folder, item):
"""画像の推論."""
if folder not in ['train', 'test']:
abort(404)
filename_list = sorted(
glob.glob(os.path.join(DATA_PATH, folder, item, '*.jpeg')))
image_list = []
for filename in filename_list:
face = Image.open(filename)
face = face.resize((IMG_ROWS, IMG_COLS), Image.LANCZOS)
face = face.convert('L')
face = np.array(face, dtype=np.float32) / 255.0
face = np.ravel(face)
image_list.append(face)
percent_list = face_deep.predict(image_list, dtype='int')
rows = []
for filename, percent in zip(filename_list, percent_list):
color = CLASSES.index(item) in [
index for index, value in enumerate(percent)
if value == max(percent)
]
row = {
'filename': os.path.basename(filename),
'percent': percent,
'color': color
}
rows.append(row)
return render_template('predict.html',
folder=folder,
item=item,
headers=CLASSES,
rows=rows)
from serial.tools import list_ports
import numpy as np
import os
from config import CLASSES
from random import choice
import uuid
app = QApplication(sys.argv)
INIT = 0
CONNECTED = 1
STARTED = 2
pixmaps = {
k: QPixmap(v).scaled(128, 128, QtCore.Qt.KeepAspectRatio)
for k, v in CLASSES.items()
}
class DataThread:
def __init__(self):
self._running = True
def terminate(self):
self._running = False
def run(self, controller, port):
while self._running:
if port.in_waiting > 0:
serialString = port.readline()
new_data = np.array(
classes_selector = st.sidebar.multiselect('Select classes',
CLASSES, default='person')
all_labels_chbox = st.sidebar.checkbox('All classes', value=False)
# Prediction section
# ---------------------------------------------------------
#target labels and their colors
#target_class_ids - индексы выбранных классов согласно списку классов MS COCC
#rgb_colors - rgb-цвета для выбранных классов
if all_labels_chbox:
target_class_ids = list(range(len(CLASSES)))
elif classes_selector:
target_class_ids = [CLASSES.index(class_name) for class_name in classes_selector]
else:
target_class_ids = [0]
rgb_colors = get_colors(target_class_ids)
detected_ids = None
if prediction_mode == 'Single image':
# добавляет форму для загрузки изображения
uploaded_file = st.file_uploader(
"Choose an image",
type=['png', 'jpg', 'jpeg'])
# если файл загружен