def __init__(self, cam, **kwargs): super(Screen, self).__init__(**kwargs) self._updateTime = 0 self._fps = "FPS:00.00" self._cam = cam self._looper = None self._classifier = RankClassifier() self._disp_mode = 'rectangle'
def __init__(self, cam, sws, **kwargs):
super(Screen, self).__init__(**kwargs)
self._updateTime = 0
self._fps = "FPS:00.00"
self._cam = cam
self._label1 = False
self._looper = None
self._classifier = RankClassifier()
self._sws = sws
self._wm = WorkloadMonitor()
class PredictionScreen(Screen): _fps = StringProperty() def __init__(self, cam, **kwargs): super(Screen, self).__init__(**kwargs) self._updateTime = 0 self._fps = "FPS:00.00" self._cam = cam self._looper = None self._classifier = RankClassifier() self._disp_mode = 'rectangle' def on_enter(self): self._looper = Clock.schedule_interval(self.on_loop, 0.1) def on_leave(self): self._looper.cancel() def on_loop(self, dt): self.capture() def capture(self): result = self._cam.capture() if not result.moving and len(result.images) > 0: h_ratio = 0 * MAX_RESIZE_RATIO #TODO: show conpane w_ratio = 0 * MAX_RESIZE_RATIO predictions = self._classifier.predict(result.images, result.areas, h_ratio, w_ratio) if self._disp_mode == 'rectangle': self.draw_box_and_label(result.images, result.rects, result.centers, predictions) else: self.draw_shadow(result.shadow, result.centers) self.draw_label(result.images, result.rects, result.centers, predictions) else: if self._disp_mode == 'rectangle': self.draw_box(result.rects) else: self.draw_shadow(result.shadow, result.centers) self.draw_fps() def draw_box_and_label(self, images, rects, centers, predictions): canvas = self.ids.monitor.canvas canvas.clear() for img, rect, c, p, acc in zip(images, rects, centers, predictions.get_all_labels(), predictions.get_all_accuraces()): #Color if p[0] < 5: color = (0.75, 0.40, 0.62, 1) elif p[0] >= 5 and p[0] < 9: color = (0.40, 0.75, 0.62, 1) elif p[0] >= 8: color = (0.7, 0.7, 0.7, 1) r,g,b,_ = color label_text = LABEL[p[0]] label = CoreLabel(text="Rank :%s"%(label_text), font_size=30, color=color, italic=True) label.refresh() texture = label.texture texture_size = list(texture.size) h,w,_ = img.shape cm = h * PX2CM_FACTOR length = CoreLabel(text="Length:%.1fcm"%(cm), font_size=30, color=color, italic=True) length.refresh() texture2 = length.texture texture2_size = list(texture2.size) accuracy = CoreLabel(text="Accuracy:%d"%(int(acc[0]*100))+"%", font_size=30, color=color, italic=True) accuracy.refresh() texture3 = accuracy.texture texture3_size = list(texture3.size) label2_text = LABEL[p[1]] label2 = CoreLabel(text="(%s)"%(label2_text), font_size=20,italic=True) label2.refresh() texture4 = label2.texture texture4_size = list(texture4.size) with canvas: Color(r, g, b) Line(points=rect, width=3, close=True) Line(points=[(c[0]-5, 110), (c[0]-35, 5), (c[0] + texture2_size[0], 5)], width=1, color=color) Rectangle(texture=texture, pos=(c[0], texture3_size[1]*2+5), size=texture_size) Rectangle(texture=texture3, pos=(c[0]-7 , texture2_size[1]+5), size=texture3_size) Rectangle(texture=texture2, pos=(c[0]-14, 5), size=texture2_size) Color(0.7,0.7,0.7) Rectangle(texture=texture4, pos=(c[0] + texture_size[0] + 10, texture3_size[1]*2+10), size=texture4_size) def draw_box(self, rects): canvas = self.ids.monitor.canvas canvas.clear() for rect in rects: with canvas: Color(0.5, 0.76, 0.86) Line(points=rect, width=2, close=True) def draw_shadow(self, shadow, centers): canvas = self.ids.monitor.canvas canvas.clear() with canvas: texture = Texture.create(size=(shadow.shape[1], shadow.shape[0]), colorfmt='rgb') texture.blit_buffer(shadow.tostring(), bufferfmt='ubyte', colorfmt='rgb') texture_size = list(texture.size) texture.flip_vertical() Rectangle(texture=texture, pos=(0, 0), size=self.size) def draw_label(self,images, rects, centers, predictions): canvas = self.ids.monitor.canvas for img, rect, c, p, acc in zip(images, rects, centers, predictions.get_all_labels(), predictions.get_all_accuraces()): #Color if p[0] < 5: color = (0.75, 0.40, 0.62, 1) elif p[0] >= 5 and p[0] < 9: color = (0.40, 0.75, 0.62, 1) elif p[0] >= 8: color = (0.7, 0.7, 0.7, 1) r,g,b,_ = color label_text = LABEL[p[0]] label = CoreLabel(text="Rank :%s"%(label_text), font_size=30, color=color, italic=True) label.refresh() texture = label.texture texture_size = list(texture.size) h,w,_ = img.shape cm = h * PX2CM_FACTOR length = CoreLabel(text="Length:%.1fcm"%(cm), font_size=30, color=color, italic=True) length.refresh() texture2 = length.texture texture2_size = list(texture2.size) accuracy = CoreLabel(text="Accuracy:%d"%(int(acc[0]*100))+"%", font_size=30, color=color, italic=True) accuracy.refresh() texture3 = accuracy.texture texture3_size = list(texture3.size) label2_text = LABEL[p[1]] label2 = CoreLabel(text="(%s)"%(label2_text), font_size=20,italic=True) label2.refresh() texture4 = label2.texture texture4_size = list(texture4.size) with canvas: Color(r, g, b) Line(points=[(c[0]-5, 110), (c[0]-35, 5), (c[0] + texture2_size[0], 5)], width=1, color=color) Rectangle(texture=texture, pos=(c[0], texture3_size[1]*2+5), size=texture_size) Rectangle(texture=texture3, pos=(c[0]-7 , texture2_size[1]+5), size=texture3_size) Rectangle(texture=texture2, pos=(c[0]-14, 5), size=texture2_size) Color(0.7,0.7,0.7) Rectangle(texture=texture4, pos=(c[0] + texture_size[0] + 10, texture3_size[1]*2+10), size=texture4_size) def draw_fps(self): elapsed = time.time() - self._updateTime fps = 1.0 / elapsed self._fps = "FPS:%02.2f"%(fps) self._updateTime = time.time() def changed_disp_mode(self, sw): if sw == 0: self._disp_mode = 'rectangle' else: self._disp_mode = 'shadow'
def main():
start_time = time.time()
# Read documents, divide according to the topics and separate train and test data-set.
t_path = os.getcwd() + "/bbc/"
all_docs = defaultdict(lambda: list())
topic_list = list()
print("Reading all the documents...\n")
for topic in os.listdir(t_path):
d_path = t_path + topic + '/'
topic_list.append(topic)
temp_docs = list()
for f in os.listdir(d_path):
f_path = d_path + f
temp_docs.append(Document(f_path, topic))
all_docs[topic] = temp_docs[:]
fold_count = 10
train_docs, test_docs = list(), list()
for key, value in all_docs.items():
random.shuffle(value)
test_len = int(len(value) / fold_count)
train_docs += value[:-test_len]
test_docs += value[-test_len:]
# Create tfidf and tfidfie index of training docs, and store into the docs.
index = Index(train_docs)
print("Train Document Count: " + str(len(train_docs)))
print("Test Document Count: " + str(len(test_docs)))
test_topics = [d.topic for d in test_docs]
for doc in train_docs:
doc.vector = doc.tfidfie
for doc in test_docs:
doc.vector = doc.tf
# create classifier instances.
nb = NaiveBayes()
rc = RankClassifier()
kmeans = KMeans(topic_list)
classifier_list = [rc, nb, kmeans]
for i in range(len(classifier_list)):
print("\nClassifier #" + str(i + 1) + "\n")
classifier = classifier_list[i]
classifier.confusion_matrix, c_dict = init_confusion_matrix(topic_list)
print("Training...\n")
classifier.train(train_docs)
print("Testing... Classifying the test docs...\n")
predictions = classifier.classify(test_docs)
# Update the confusion matrix and statistics with updated values.
classifier.confusion_matrix = update_confusion_matrix(
test_topics, predictions, classifier.confusion_matrix, c_dict)
classifier.stats = cal_stats(classifier.confusion_matrix)
print("Confusion Matrix\n")
for item in classifier.confusion_matrix:
print(item)
print("\nStatistics\n")
print_table(get_stats_table(classifier.stats))
print("Run time...{} secs \n".format(round(time.time() - start_time, 4)))
# call recommendation system once classifiers are ready.
recommendation(all_docs, test_docs, classifier_list)
def main():
start_time = time.time()
t_path = "../data_set/bbc/"
all_docs = defaultdict(lambda: list())
topic_list = list()
print("Reading all the documents...\n")
print(os.listdir(t_path))
for topic in os.listdir(t_path):
d_path = t_path + topic + '/'
topic_list.append(topic)
temp_docs = list()
for f in os.listdir(d_path):
f_path = d_path + f
temp_docs.append(Document(f_path, topic))
all_docs[topic] = temp_docs[:]
fold_count = 10
train_docs, test_docs = list(), list()
for key, value in all_docs.items():
random.shuffle(value)
test_len = int(len(value) / fold_count)
train_docs += value[:-test_len]
test_docs += value[-test_len:]
index = Index(train_docs)
print("Train Document Count: " + str(len(train_docs)))
print("Test Document Count: " + str(len(test_docs)))
test_topics = [d.topic for d in test_docs]
for doc in train_docs:
doc.vector = doc.tfidfie
for doc in test_docs:
doc.vector = doc.tf
nb = NaiveBayes()
rc = RankClassifier()
kmeans = KMeans(topic_list)
classifier_list = [rc, nb, kmeans]
for i in range(len(classifier_list)):
print("\nClassifier #" + str(i + 1) + "\n")
classifier = classifier_list[i]
classifier.confusion_matrix, c_dict = init_confusion_matrix(topic_list)
print("Training...\n")
classifier.train(train_docs)
print("Testing... Classifying the test docs...\n")
predictions = classifier.classify(test_docs)
classifier.confusion_matrix = update_confusion_matrix(
test_topics, predictions, classifier.confusion_matrix, c_dict)
classifier.stats = cal_stats(classifier.confusion_matrix)
print("Confusion Matrix\n")
for item in classifier.confusion_matrix:
print(item)
print("\nStatistics\n")
print_table(get_stats_table(classifier.stats))
print("Run time...{} secs \n".format(round(time.time() - start_time, 4)))
recommendation(all_docs, test_docs, classifier_list)
def main():
start_time = time.time()
# Read documents, divide according to the topics and separate train and test data-set.
t_path = "../bbc/"
all_docs = defaultdict(lambda: list())
topic_list = list()
for topic in os.listdir(t_path):
d_path = t_path + topic + '/'
topic_list.append(topic)
temp_docs = list()
for f in os.listdir(d_path):
f_path = d_path + f
temp_docs.append(Document(f_path, topic))
all_docs[topic] = temp_docs[:]
fold_count = 10
train_docs, test_docs = list(), list()
for key, value in all_docs.items():
random.shuffle(value)
test_len = int(len(value) / fold_count)
train_docs += value[:-test_len]
# explanation
# lis = [1,2,3,4,5]
# print(lis[:-4])
# print(lis[-4:])
test_docs += value[-test_len:]
# Create tfidf and tfidfie index of training docs, and store into the docs.
index = Index(train_docs)
test_topics = [d.topic for d in test_docs]
for doc in train_docs:
doc.vector = doc.tfidfie
for doc in test_docs:
doc.vector = doc.tf
# create classifier instances.
nb = NaiveBayes()
rc = RankClassifier()
kmeans = KMeans(topic_list)
classifier_list = [nb, rc, kmeans]
for i in range(len(classifier_list)):
classifier = classifier_list[i]
classifier.confusion_matrix, c_dict = init_confusion_matrix(topic_list)
classifier.train(train_docs)
predictions = classifier.classify(test_docs)
# Update the confusion matrix and statistics with updated values.
classifier.confusion_matrix = update_confusion_matrix(
test_topics, predictions, classifier.confusion_matrix, c_dict)
classifier.stats = cal_stats(classifier.confusion_matrix)
global lst
lst = []
lst.append(all_docs)
lst.append(test_docs)
lst.append(classifier_list)
return redirect('http://localhost:5000/recommend')
class PredictionScreen(Screen):
_fps = StringProperty()
def __init__(self, cam, sws, **kwargs):
super(Screen, self).__init__(**kwargs)
self._updateTime = 0
self._fps = "FPS:00.00"
self._cam = cam
self._label1 = False
self._looper = None
self._classifier = RankClassifier()
self._sws = sws
self._wm = WorkloadMonitor()
def on_enter(self):
self._looper = Clock.schedule_interval(self.on_loop, 0.3)
def on_leave(self):
self._looper.cancel()
def capture(self):
result = self._cam.capture()
if not result.moving and len(result.images) > 0:
h_ratio = self._sws.length_meter.get_balance() * MAX_RESIZE_RATIO
w_ratio = self._sws.width_meter.get_balance() * MAX_RESIZE_RATIO
predictions = self._classifier.predict(result.images, result.areas,
h_ratio, w_ratio)
self.draw_box_and_label(result.images, result.rects,
result.centers, predictions)
self._wm.count(predictions.get_top_labels())
else:
self.draw_box(result.rects)
if len(result.images) == 0:
self._wm.clear()
def on_loop(self, dt):
select = self._sws.select_meter.get_balance()
if select < 0:
self._label1 = False
else:
self._label1 = True
mode = self._sws.get_mode_str()
if mode == 'prediction':
self.capture()
else:
self.manager.current = mode
def draw_box_and_label(self, images, rects, centers, predictions):
elapsed = time.time() - self._updateTime
fps = 1.0 / elapsed
canvas = self.ids.monitor.canvas
canvas.clear()
for img, rect, c, p, acc in zip(images, rects, centers,
predictions.get_all_labels(),
predictions.get_all_accuraces()):
#Color
if p[0] < 5:
color = (0.75, 0.40, 0.62, 1)
elif p[0] >= 5 and p[0] < 9:
color = (0.40, 0.75, 0.62, 1)
elif p[0] >= 8:
color = (0.7, 0.7, 0.7, 1)
r, g, b, _ = color
label_text = LABEL1[p[0]] if self._label1 else LABEL2[p[0]]
label = CoreLabel(text="Rank :%s" % (label_text),
font_size=30,
color=color,
italic=True)
label.refresh()
texture = label.texture
texture_size = list(texture.size)
h, w, _ = img.shape
cm = h * PX2CM_FACTOR
length = CoreLabel(text="Length:%.1fcm" % (cm),
font_size=30,
color=color,
italic=True)
length.refresh()
texture2 = length.texture
texture2_size = list(texture2.size)
accuracy = CoreLabel(text="Accuracy:%d" % (int(acc[0] * 100)) +
"%",
font_size=30,
color=color,
italic=True)
accuracy.refresh()
texture3 = accuracy.texture
texture3_size = list(texture3.size)
label2_text = LABEL1[p[1]] if self._label1 else LABEL2[p[1]]
label2 = CoreLabel(text="(%s)" % (label2_text),
font_size=20,
italic=True)
label2.refresh()
texture4 = label2.texture
texture4_size = list(texture4.size)
with canvas:
Color(r, g, b)
Line(points=rect, width=3, close=True)
Line(points=[(c[0] - 5, 110), (c[0] - 35, 5),
(c[0] + texture2_size[0], 5)],
width=1,
color=color)
Rectangle(texture=texture,
pos=(c[0], texture3_size[1] * 2 + 5),
size=texture_size)
Rectangle(texture=texture3,
pos=(c[0] - 7, texture2_size[1] + 5),
size=texture3_size)
Rectangle(texture=texture2,
pos=(c[0] - 14, 5),
size=texture2_size)
Color(0.7, 0.7, 0.7)
Rectangle(texture=texture4,
pos=(c[0] + texture_size[0] + 10,
texture3_size[1] * 2 + 10),
size=texture4_size)
self._fps = "FPS:%02.2f" % (fps)
self._updateTime = time.time()
def draw_box(self, rects):
elapsed = time.time() - self._updateTime
fps = 1.0 / elapsed
canvas = self.ids.monitor.canvas
canvas.clear()
for rect in rects:
with canvas:
Color(0.5, 0.76, 0.86)
Line(points=rect, width=2, close=True)
self._fps = "FPS:%02.2f" % (fps)
self._updateTime = time.time()