def test_image(args, model):
if args.detector == 'dlib':
import dlib
elif args.detector == 'faceboxes':
from utils.face_detector import FaceDetectorFaceboxes
model.eval()
device = torch.device("cuda" if args.gpu else "cpu")
image = Image.open(args.image).convert('RGB')
if args.resize > 0:
image = resize(image, args.resize)
detector = None
if args.detector == 'dlib':
detector = dlib.get_frontal_face_detector()
elif args.detector == 'faceboxes':
MODEL_PATH = 'model/faceboxes.pb'
detector = FaceDetectorFaceboxes(MODEL_PATH, gpu_memory_fraction=0.25, visible_device_list='0')
segmenter = Segmenter(model, device, detector, mode=args.detector)
result = segmenter.segment(PIL2opencv(image), args.remove_small_area)
result = opencv2PIL(result)
if args.save:
result.save(args.save)
if not args.unshow:
result.show()
image.show()
def test_video(args, model):
if args.video == '0':
cap = cv2.VideoCapture(0)
else:
cap = cv2.VideoCapture(args.video)
w_win = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h_win = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
print(w_win, h_win)
if args.resize > 0:
short_size = args.resize
if w_win > h_win:
nw, nh = short_size, int(w_win * short_size / h_win)
else:
nw, nh = int(h_win * short_size / w_win), short_size
else:
nw, nh = w_win, h_win
detector = None
if args.detector == 'dlib':
detector = dlib.get_frontal_face_detector()
elif args.detector == 'faceboxes':
MODEL_PATH = 'model/faceboxes.pb'
detector = FaceDetectorFaceboxes(MODEL_PATH, gpu_memory_fraction=0.25, visible_device_list='0')
device = torch.device("cuda" if args.gpu else "cpu")
segmenter = Segmenter(model, device, detector, mode=args.detector)
if args.save:
fourcc = cv2.VideoWriter_fourcc(*'MJPG')
out = cv2.VideoWriter(args.save, fourcc, 20, (nh, nw), True)
while True:
frame = cap.read()[1]
if frame is None:
break
frame = cv2.resize(frame, (nh, nw))
result = segmenter.segment(frame, args.remove_small_area)
if args.save:
out.write(result)
if not args.unshow:
cv2.imshow('image', result)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()
if args.save:
out.release()
if mode == 1:
yield (X_batch, y_batch)
else:
yield X_batch
# очищаем матрицы порции для новой порции
X_batch.fill(0)
y_batch.fill(0)
batch_index = 0
w2v = gensim.models.KeyedVectors.load_word2vec_format(
word2vector_path, binary=not word2vector_path.endswith('.txt'))
w2v_dims = len(w2v.syn0[0])
segmenter = Segmenter()
tokenizer = Tokenizer()
print('Collecting samples...')
samples = []
all_words = set([PAD_WORD])
max_phrase_len = 0
if True:
# добавляем пары предпосылка-вопрос из обучающего датасета
with codecs.open(os.path.join(data_folder, qa_path), "r", "utf-8") as inf:
loading_state = 'T'
text = []
sys.stdout.encoding).strip().lower()
phrases1 = []
segm_mode = raw_input(
'Use EOL markers (1) or segmenter (2) to split file to sentences?'
).strip()
max_nb_facts = int(
raw_input(
'maximum number of samples to read from file (-1 means all):\n> ').
strip())
if max_nb_facts == -1:
max_nb_facts = 10000000
if segm_mode == 2:
segmenter = Segmenter()
phrases0 = segmenter.split(
codecs.open(path1, 'r', 'utf-8').readlines())
for phrase in enumerate(phrases):
words = tokenizer.tokenize(phrase)
if len(words) > 0:
phrases1.append(words)
if len(phrases1) >= max_nb_facts:
break
else:
with codecs.open(path1, 'r', 'utf-8') as rdr:
for phrase in rdr:
words = tokenizer.tokenize(phrase)
if len(words) > 0:
phrases1.append(words)
if len(phrases1) >= max_nb_facts:
# Created on: 00:59:12
import os
import sys
import numpy as np
import gensim
from copy import deepcopy
sys.path.append(
os.path.join(os.path.dirname(os.path.realpath(__file__)), os.pardir))
from config import Config
from utils import log
from utils.segmenter import Segmenter
from utils.tfidf import word_idf, sif_embedding, get_weighted_average
from utils.tools import load_embedding, cosine
config = Config()
cut = Segmenter()
logger = log.getLogger(__name__)
class Embedding(object):
def __init__(self, embedding_path, documents, data_seg):
# default pre-trained word embedding
self._word_embedding = gensim.models.KeyedVectors.load_word2vec_format(
embedding_path, binary=True)
# Topic clusters documents
self.documents = documents
self.data_seg = data_seg
self.weights_of_words = word_idf(self.documents)
# initized some components for sentence embedding
self.word2idx, self.wv_mat = None, None
self._word2vec()