def __init__(self):
'''
Main module that ties all the other relevant models (Segmentor, BoundingBox, Classifier)
'''
# GLOBAL CONSTANTS
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu') # Device configuration
# model paths
CLASSIFIER_DIR = 'saved_models'
CLASSIFIER_NAME = 'inference_resnet18_layers_6_opt_SGD_lr_0.0001_wd_0.1.pth'
classifier_path = os.path.join(CLASSIFIER_DIR, CLASSIFIER_NAME)
SEG_DIR = 'saved_models'
SEG_PATH = 'AUGGI_SEGNET_STATE_2018_11_28_5_7_NEPOCHS_100_TRAINAVGLOSS_13_8_TESTAVGLOSS_13_7.pth'
seg_path = os.path.join(SEG_DIR, SEG_PATH)
# ------------- Instantiate Models ------------- #
# Instantiate Seg, Boxer and Classifier models
self.segmentor = Segmentor(seg_path)
self.boxer = BoundingBox()
self.classifier = Classifier(classifier_path)
def seg_eval(args):
if not (DatasetHandler.is_readable(args.gold_file)):
logging.error("path '%s' open failed !" % (args.gold_file))
logging.error('Exit!')
exit(1)
if not DatasetHandler.is_readable(args.predict_file):
logging.error("path '%s' open failed ! predict file open error ." %
(args.predict_file))
logging.error("Exit!")
exit(1)
segmentor = Segmentor()
segmentor.evaluate(args.gold_file, args.predict_file)
def seg_train(args):
if not DatasetHandler.is_readable(args.training_file):
logging.error("path '%s' open failed !" % (args.training_file))
logging.error('Exit!')
exit(1)
if not DatasetHandler.is_readable(args.developing_file):
logging.error("path '%s' open failed !" % (args.developing_file))
logging.error("Exit!")
exit(1)
if not DatasetHandler.is_writeable(args.model_saving):
logging.error("path '%s' open failed !" % (args.model_saving))
logging.error('Exit!')
exit(1)
segmentor = Segmentor()
segmentor.train(args.training_file, args.developing_file,
args.model_saving, args.max_iter)
def seg_predict(args):
if not DatasetHandler.is_readable(args.predict_file):
logging.error("path '%s' open failed !" % (args.predict_file))
logging.error('Exit!')
exit(1)
if not DatasetHandler.is_readable(args.model_loading):
logging.error("path '%s' open failed ! Model load Error ." %
(args.model_loading))
logging.error("Exit!")
exit(1)
if not DatasetHandler.is_writeable(
args.output_path) and args.output_path != "stdout":
logging.error("path '%s' open failed !" % (args.output_path))
logging.error('Exit!')
exit(1)
segmentor = Segmentor()
segmentor.predict(args.model_loading, args.predict_file, args.output_path)
def mask_and_save_silhouettes(base_path):
if not os.path.exists(os.path.join(
base_path, RAW_SILHOUETTES_DIR_NAME)) and os.path.exists(
os.path.join(base_path, IMAGE_DIR_NAME)):
image_paths = filepath_at_dir_sorted(
os.path.join(base_path, IMAGE_DIR_NAME), sorter)
if len(image_paths) > 0:
os.makedirs(os.path.join(base_path, RAW_SILHOUETTES_DIR_NAME))
segment = Segmentor()
try:
for path in image_paths:
segment.segment_images_at_path(
path,
os.path.join(base_path, RAW_SILHOUETTES_DIR_NAME, ''))
except Exception as e:
if os.path.exists(
os.path.join(base_path, RAW_SILHOUETTES_DIR_NAME)):
shutil.rmtree(
os.path.join(base_path, RAW_SILHOUETTES_DIR_NAME))
raise e
silhouettes, file_paths = load_images_of_type(base_path,
RAW_SILHOUETTES_DIR_NAME)
if len(silhouettes) > 0:
os.makedirs(os.path.join(base_path, SILHOUETTES_DIR_NAME))
index = 1
for sil in silhouettes:
sil = np.array(sil)
new_sil = np.zeros([sil.shape[0], sil.shape[1]])
mask = lambda im: im[:, :] > 0
if sil.shape[2] > 1:
new_sil[mask(sil[:, :, 0]) | mask(sil[:, :, 1])
| mask(sil[:, :, 2])] = 1
save_np_data(os.path.join(base_path, SILHOUETTES_DIR_NAME),
'%d.npy' % index, new_sil)
index += 1
def test_segmentor(self):
tests = [
([1, 1, 0, 1, 0, 0, 1, 0, 1], [[0, 1, 2, 3]]),
([1, 1, 0, 1, 1, 0, 1, 0, 1], [[0, 1, 2, 3, 4, 5, 6]]),
([1, 1, 0, 1, 1, 0, 1, 1, 1], [list(range(0, 8 + 1))]),
([1, 1, 1, 0, 1], [[0, 1, 2, 3, 4]]),
([0, 0, 0, 0, 0], []),
([1] * 7 + [0] * 3, [list(range(7))])
]
min_frames = 4
threshold = 0.7
for ex, exp in tests:
self.assertEqual(exp, Segmentor._segmentor(ex, min_frames, threshold))
class Inference:
def __init__(self):
'''
Main module that ties all the other relevant models (Segmentor, BoundingBox, Classifier)
'''
# GLOBAL CONSTANTS
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu') # Device configuration
# model paths
CLASSIFIER_DIR = 'saved_models'
CLASSIFIER_NAME = 'inference_resnet18_layers_6_opt_SGD_lr_0.0001_wd_0.1.pth'
classifier_path = os.path.join(CLASSIFIER_DIR, CLASSIFIER_NAME)
SEG_DIR = 'saved_models'
SEG_PATH = 'AUGGI_SEGNET_STATE_2018_11_28_5_7_NEPOCHS_100_TRAINAVGLOSS_13_8_TESTAVGLOSS_13_7.pth'
seg_path = os.path.join(SEG_DIR, SEG_PATH)
# ------------- Instantiate Models ------------- #
# Instantiate Seg, Boxer and Classifier models
self.segmentor = Segmentor(seg_path)
self.boxer = BoundingBox()
self.classifier = Classifier(classifier_path)
# ---------------------------------------------- #
def predict(self, image):
# ------------ Run full inference on image ----- #
binary_mask = self.segmentor.segment(image) # get binary mask (sending rgb PIL)
x_min, y_min, x_max, y_max = self.boxer.get_box(binary_mask) # get bbox, sending grayscale PIL image
# classify the (original) image, pass bboxes, receive a bristol prediction
bristol_pred = self.classifier.classify(image, x_min, y_min, x_max, y_max)
return bristol_pred
from naivebayes import NaiveBayesClassifier
import os
import re
import codecs
from segmentor import Segmentor
def corpus_generator(segmentor):
for corpus in map(lambda x: "sentiment_corpus/" + x, ["Ctrip_htl_ba_4000", "Dangdang_Book_4000", "Jingdong_NB_4000"]):
classes = filter(lambda x: x[0] != ".", os.listdir(corpus))
for cls in classes:
print "Enumerating for '%s/%s' reviews." % (corpus, cls)
cls_dir = os.path.join(corpus, cls)
files = filter(lambda x: x.endswith(".txt"), os.listdir(cls_dir))
for filename in files:
with codecs.open(os.path.join(cls_dir, filename), "r", encoding="utf8") as file:
for line in file:
if not line.strip():
continue
words = segmentor(line.strip())
yield (cls, words)
segmentor = Segmentor()
generator = corpus_generator(segmentor)
classifier = NaiveBayesClassifier()
classifier.train(generator)
print classifier.classify(segmentor(u"这一地区生鲜奶收购价持续在低位徘徊，导致很多奶户入不敷出，被迫“砍牛”（杀牛或卖牛）。 近期，双鸭山市多地奶农联名向记者反映"))
# print classifier.classify("This is awesome but still I don't like it thisisaweirdwordneveroccurs. ".split(" "))
# print classifier.classify("iqbvajkkjbarjta".split(" "))
# print classifier.classify("I don't recommend.".split(" "))
images_path = list(list_files())
#image_path = images_path[3]
# image_path = "dataset/pes1.png"
times = []
set_stats = []
for image_path in images_path:
image = cv2.imread(image_path)
image = cv2.resize(image, (512, 512))
image_gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
segmentor = Segmentor(image_gray)
duration, out = segmentor.max_flow_gray()
rect = get_rect(out)
segmentor = Segmentor_grab(image)
duration1, out1 = segmentor.segment(rect)
out1 = cv2.cvtColor(out1, cv2.COLOR_BGR2GRAY)
set_stats.append(compare_images(out, out1))
times.append((duration, duration1))
output_path = OUTPUT_DIR + image_path.split('/')[1].split(
'.')[0] + "_graph_cut.jpg"
output_path1 = OUTPUT_DIR + image_path.split('/')[1].split(
'.')[0] + "_grab_cut.jpg"