def main():
words = load_words()
sess = tf.InteractiveSession()
path = PATH.DATASET_DIR+'test7x1x2.record'
#iterator = read_record(path, epochs=10, batch_size=1024)
iterator = read_from_npy(batch_size=1024)
x, y = iterator.get_next()
y = tf.cast(y, tf.int64)
x = tf.reshape(x, (-1, 48,48,1))
y_in = tf.one_hot(y, 3557, dtype=tf.int64)
y_out = built_net(x)
train_op, loss, acc = optimizer(y_in, y_out)
tf.global_variables_initializer().run()
t0 = time.time()
for epoch in range(1, 100):
for i in range(235):
sess.run(train_op)
if i%10 == 0:
x_,y_ = sess.run([y,y_out])
print('in: %s\nout: %s' %(x_[1:10], np.argmax(y_, 1)[1:10]))
losses, accurary = sess.run([loss, acc])
print('[==>] Epoch: %d \tStep: %d \tLoss: %s \tAcc: %s \tTime: %ss'
%(epoch, i, losses, accurary, round(time.time()-t0, 2)))
t0 = time.time()
def predict(imgs, model, xy, save=False, cls=None):
img01 = imgs / 255
#print(imgs)
if cls == None:
cls = load_words()
st = time.time()
predicts = model.predict(img01)
result = ''
new_xy = []
for i, probabilities in enumerate(predicts):
index_cls = np.argmax(probabilities)
if index_cls == 3557:
print(index_cls)
if index_cls == 3557 or probabilities[index_cls] < 0.8:
continue
new_xy.append(xy[i])
result += cls[index_cls]
if save:
nrootdir = ("./cut_image/")
if not os.path.isdir(nrootdir):
os.makedirs(nrootdir)
cv2.imwrite(nrootdir + cls[index_cls] + ".jpg", imgs[i])
#print('Result: %s' % (result))
et = time.time()
#print('花费时间: %s' % (et - st))
return result, new_xy
def get_frame_from_camera():
print('Get serial...')
p = serial.tools.list_ports.comports()[0]
serialName = p[0]
serialFd = serial.Serial(serialName, 9600, timeout=60)
print('Loading classes...')
cls = load_words()
print('Loading model...')
#model_path = PATH.MODEL_DIR + 'model28x3_1.h5'
model_path = PATH.MODEL_DIR + 'model_keras/model.h5'
model = keras.models.load_model(model_path)
camera = cv2.VideoCapture(0)
camera.set(3, 640)
camera.set(4, 480)
serialFd.write(b'@PlayFlashText#099$')
state = RUN
while True:
_, frame = camera.read()
w, h, c = frame.shape
s = int(part_num / 2)
w, h = int(w / part_num), int(h / part_num)
focus_frame = frame[w * s:w * (s + 1), h * s:h * (s + 1), :]
processed_frame, regions, xy = findContour(focus_frame)
regions = np.array(regions)
show_frame = draw_frame(serialFd, processed_frame, regions, xy, model,
cls)
state = controler(state, serialFd)
show_frame = action(state, show_frame)
#cv2.rectangle(show_frame,(w*s, h*s),(w*(s+1), h*(s+1)),(255, 0, 0))
cv2.imshow('processed frame', show_frame)
key = cv2.waitKey(1)
if key == ord('q'):
break
camera.release()
cv2.destroyAllWindows()
def predict(data, model, save=True, cls=None):
im = np.round(data / 255.)
if cls == None:
cls = load_words()
st = time.time()
predicts = model.predict(im)
result = ''
for i, probabilities in enumerate(predicts):
index = np.argmax(probabilities)
if probabilities[index] < 0.01:
continue
result += cls[index]
if save:
nrootdir = ("./pic/")
if not os.path.isdir(nrootdir):
os.makedirs(nrootdir)
cv2.imwrite(nrootdir + cls[index] + ".jpg", data[i])
print('Result: %s' % (result))
et = time.time()
print('花费时间: %s' % (et - st))
return result
def main():
cls = load_words()
model_path = PATH.MODEL_DIR + 'model28x3_1.h5'
print('Loading model...')
model = K.models.load_model(model_path)
train_font_dir = PATH.TRAIN_FONT_DIR
fonts = read_fontnames(train_font_dir)
while True:
words = input('请输入要生成的图片文字\n')
data = gene_data(words, [fonts[3]], train_font_dir)
images = []
for image in np.array(data):
kernel = np.ones((1, 1), np.uint8)
erosion = cv2.erode(image, kernel)
kernel = np.ones((2, 2), np.uint8)
dilation = cv2.dilate(erosion, kernel)
dilation = np.round(dilation, 0)
cv2.imshow('hh', dilation)
cv2.waitKey(0)
images.append(dilation)
images = np.array(images).reshape(-1, 48, 48, 1)
predict(images, model, cls=cls)
import keras
from configuration.config import PATH
model_name = 'model1/model2.h5'
model = keras.models.load_model(PATH.MODEL_DIR + model_name)
import cv2
import numpy as np
import utils.utils as tool
im = cv2.imread('test_pic/13.jpg')
im = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)
im = im.reshape([1, 48, 48, 1])
r = model.predict(im)
r = np.argmax(r)
cls = tool.load_words()
print(cls[r])