def classify_image(self, string):
img = snapshot()
self.img_copy = img.resize(224, 224)
a = self.img_copy.pix_to_ai()
a = img.draw_string(0,
rows[self.row],
string,
color=(255, 255, 255),
scale=1.5,
mono_space=False)
a = lcd.display(img)
del (img)
gc.collect()
def detect_objects(self, threshold, return_img=False):
detected = False
img = snapshot()
img_copy = img.resize(224, 224)
a = img_copy.pix_to_ai()
code = kpu.run_yolo2(self.object_detection_task, img_copy)
if code:
for i in code:
if i.value() >= threshold:
detected = True
new_x, new_y = int(i.x() * 1.07), int(i.y() * 1.42)
roi = (new_x, new_y, int(i.w() * 1.07), int(i.h() * 1.42))
percent = i.value()
object_detected = self.classes[i.classid()]
if not return_img:
a = img.draw_rectangle(roi,
color=(0x1c, 0xa2, 0xff),
thickness=2)
a = img.draw_string(new_x,
new_y - 14,
("%s %%: %.2f" %
(object_detected, percent)),
color=(255, 255, 255),
scale=1.5,
mono_space=False)
self.x_center, self.y_center, self.area, self.percent, roi = _find_max(
code)
del (img_copy)
if not detected:
self.object_detected = None
self.percent = -1
if return_img:
return img, None
if return_img:
return img, roi
else:
a = lcd.display(img)
del (img)
def classify_image(self, threshold):
img = snapshot()
img_copy = img.resize(224, 224)
a = img_copy.pix_to_ai()
fmap = kpu.forward(self.image_objects_task, img_copy)
plist = fmap[:]
pmax = max(plist)
self.percent = pmax
if self.percent >= threshold:
max_index = plist.index(pmax)
a = img.draw_string(8,
rows[self.row],
("Result: %s %%: %.2f" %
(self.labels[max_index].strip(), pmax)),
color=(255, 255, 255),
scale=1.5,
mono_space=False)
self.image_class = self.labels[max_index].strip()
a = lcd.display(img)
del (img)
del (img_copy)
def class_predict(self, THRESHOLD=50):
img = snapshot()
self.img_copy = img.resize(224, 224)
#kpu.memtest()
a = self.img_copy.pix_to_ai()
res_index = -1
try:
res_index, min_dist = self.classifier.predict(self.img_copy)
self.percent = round(min_dist, 0)
#print(" percent:",self.percent)
#print("res_index: ", res_index)
#print("{:.2f}".format(min_dist))
except Exception as e:
print("predict err:", e)
if res_index >= 0 and self.percent >= THRESHOLD:
self.image_class = self.categoryname[self.class_list[res_index]]
#print("predict result:", self.image_class)
a = lcd.display(img)
del (img)
#del(img_copy)
gc.collect()
def record_samples_training(self):
self.classify_image(("Samples taken: %d" % (self.cap_num - 3)))
if self.cap_num < self.classnumber + self.samplesnumber:
index = self.classifier.add_sample_img(self.img_copy)
self.cap_num += 1
if self.train_status == 0:
if self.cap_num >= self.classnumber + self.samplesnumber:
print("start train")
img = snapshot()
a = img.draw_string(lcd.width() // 2 - 68,
lcd.height() // 2 - 4,
"Training...",
color=(0, 255, 0),
scale=3,
mono_space=False)
lcd.display(img)
del (img)
#del(self.img_copy)
gc.collect()
time.sleep(2)
self.classifier.train()
print("train end")
self.train_status = 1