def facial_detact(modelPath, showBox=True):
face_x = tf.placeholder(tf.float32, [None, 2304])
y_conv = deepnn(face_x)
probs = tf.nn.softmax(y_conv)
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(modelPath)
sess = tf.Session()
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
print(
'Restore model sucsses!!\n\nNOTE: Press SPACE on keyboard to capture face.'
)
feelings_faces = []
for index, emotion in enumerate(EMOTIONS):
feelings_faces.append(
cv2.imread('./data/emojis/' + emotion + '.png', -1))
video_captor = cv2.VideoCapture(0)
emoji_face = []
result = None
while True:
ret, frame = video_captor.read()
detected_face, face_coor = format_image(frame)
if showBox:
if face_coor is not None:
[x, y, w, h] = face_coor
cv2.rectangle(frame, (x, y), (x + w, y + h), (0, 0, 255), 2)
if cv2.waitKey(1) & 0xFF == ord(' '):
if detected_face is not None:
cv2.imwrite('a.jpg', detected_face)
tensor = image_to_tensor(detected_face)
result = sess.run(probs, feed_dict={face_x: tensor})
print(EMOTIONS[np.argmax(result[0])])
if result is not None:
for index, emotion in enumerate(EMOTIONS):
cv2.putText(frame, emotion, (10, index * 20 + 20),
cv2.FONT_HERSHEY_PLAIN, 0.5, (0, 255, 0), 1)
cv2.rectangle(frame, (130, index * 20 + 10),
(130 + int(result[0][index] * 100),
(index + 1) * 20 + 4), (255, 0, 0), -1)
emoji_face = feelings_faces[np.argmax(result[0])]
for c in range(0, 3):
frame[200:320, 10:130, c] = emoji_face[:, :, c] * (
emoji_face[:, :, 3] /
255.0) + frame[200:320, 10:130,
c] * (1.0 - emoji_face[:, :, 3] / 255.0)
cv2.imshow('face', frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
video_captor.release()
cv2.destroyAllWindows()
def demo(modelPath, showBox=True):
face_x = tf.placeholder(tf.float32, [None, 2304])
y_conv = deepnn(face_x)
probs = tf.nn.softmax(y_conv)
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(modelPath)
sess = tf.Session()
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
feelings_faces = []
for index, emotion in enumerate(EMOTIONS):
feelings_faces.append(
cv2.imread('./data/emojis/' + emotion + '.png', -1))
emoji_face = []
img = input("Enter the image file name: ")
while True:
frame = cv2.imread(img)
detected_face, face_coor = format_image(frame)
if face_coor is not None:
[x, y, w, h] = face_coor
cv2.rectangle(frame, (x, y), (x + w, y + h), (255, 0, 0), 2)
result = None
if detected_face is not None:
cv2.imwrite('face.jpg', detected_face)
tensor = image_to_tensor(detected_face)
result = sess.run(probs, feed_dict={face_x: tensor})
if result is not None:
for index, emotion in enumerate(EMOTIONS):
cv2.putText(frame, emotion, (10, index * 20 + 20),
cv2.FONT_HERSHEY_PLAIN, 0.5, (0, 255, 0), 1)
cv2.rectangle(frame, (130, index * 20 + 10),
(130 + int(result[0][index] * 100),
(index + 1) * 20 + 4), (255, 0, 0), -1)
emoji_face = feelings_faces[np.argmax(result[0])]
for c in range(0, 3):
frame[200:320, 10:130, c] = emoji_face[:, :, c] * (
emoji_face[:, :, 3] /
255.0) + frame[200:320, 10:130,
c] * (1.0 - emoji_face[:, :, 3] / 255.0)
else:
comment = 'We cannot detect a face....Too bad!'
cv2.putText(frame, comment, (115, 250), cv2.FONT_HERSHEY_PLAIN,
1.5, (0, 0, 0), 1)
cv2.imwrite('result1.jpg', frame)
cv2.namedWindow('Guess how I feel', cv2.WINDOW_NORMAL)
cv2.resizeWindow('Guess how I feel', 450, 600)
cv2.imshow('Guess how I feel', frame)
k = cv2.waitKey(10) & 0xff # Press 'ESC' for exiting video
if k == 27:
break
def demo(modelPath, showBox=False,SAMPLE_IMAGE_PATH=SAMPLE_IMAGE):
face_x = tf.placeholder(tf.float32, [None, 2304])
y_conv = deepnn(face_x)
probs = tf.nn.softmax(y_conv)
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(modelPath)
sess = tf.Session()
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
result = None
frame = SAMPLE_IMAGE_PATH
detected_face, face_coor = format_image(frame) #face Coordinates
if showBox:
if face_coor is not None:
[x,y,w,h] = face_coor
cv2.rectangle(frame, (x,y), (x+w,y+h), (255,0,0), 2)
if detected_face is not None:
cv2.imwrite('temp.jpg', detected_face)
tensor = image_to_tensor(detected_face)
result = sess.run(probs, feed_dict={face_x: tensor})
print(result)
jsonString='{'
maximum=0.00
FINAL_EMOTION=''
if result is not None:
for index, emotion in enumerate(EMOTIONS):
if(index!=0):
jsonString+=','
percentage= float("{0:.2f}".format(result[0][index]*100))
if(percentage==0.0):
percentage=0
jsonString+='"'+str(emotion)+'":'+str(percentage)
if(maximum<percentage):
maximum=copy.copy(percentage)
FINAL_EMOTION=emotion
print('--------------------------------')
print(' ')
print(str(emotion)+" : "+str(percentage)+"%")
print(' ')
print('=========================================================')
print('The Result for Input Image is : '+str(FINAL_EMOTION)+" ( "+str(maximum)+"% )")
print('=========================================================')
jsonString+='}'
return jsonString
def main(_):
mnist = input_data.read_data_sets(FLAGS.data_dir, one_hot=True)
with tf.variable_scope('model'):
x = tf.placeholder(tf.float32, [None, 784], name="x")
y_ = tf.placeholder(tf.float32, [None, 10], name='y_')
training = tf.placeholder(bool, (), name='mode')
ybar, logits = deepnn(x, logits=True, training=training)
cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(
labels=y_, logits=logits),
name="cse")
correct_prediction = tf.equal(tf.argmax(ybar, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32),
name="acc")
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
for i in range(FLAGS.train_steps):
batch = mnist.train.next_batch(50)
if i % 100 == 0:
train_accuracy = accuracy.eval(feed_dict={
x: batch[0],
y_: batch[1],
training: False
})
print('step %d, training accuracy %g' % (i, train_accuracy))
train_step.run(feed_dict={
x: batch[0],
y_: batch[1],
training: True
})
saver = tf.train.Saver()
saver.save(sess, FLAGS.output_dir + "/model")
print('test accuracy %g' % accuracy.eval(feed_dict={
x: mnist.test.images,
y_: mnist.test.labels,
training: False
}))
def demo(modelPath, showBox=False):
face_x = tf.placeholder(tf.float32, [None, 2304])
y_conv = deepnn(face_x)
probs = tf.nn.softmax(y_conv)
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(modelPath)
sess = tf.Session()
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
print('Restore model sucsses!!\nNOTE: Press SPACE on keyboard to capture face.')
feelings_faces = []
for index, emotion in enumerate(EMOTIONS):
feelings_faces.append(cv2.imread('./data/emojis/' + emotion + '.png', -1))
video_captor = cv2.VideoCapture(0)
emoji_face = []
result = None
while True:
ret, frame = video_captor.read()
detected_face, face_coor = format_image(frame)
if showBox:
if face_coor is not None:
[x,y,w,h] = face_coor
cv2.rectangle(frame, (x,y), (x+w,y+h), (255,0,0), 2)
if cv2.waitKey(1) & 0xFF == ord(' '):
if detected_face is not None:
cv2.imwrite('a.jpg', detected_face)
tensor = image_to_tensor(detected_face)
result = sess.run(probs, feed_dict={face_x: tensor})
# print(result)
if result is not None:
for index, emotion in enumerate(EMOTIONS):
cv2.putText(frame, emotion, (10, index * 20 + 20), cv2.FONT_HERSHEY_PLAIN, 0.5, (0, 255, 0), 1)
cv2.rectangle(frame, (130, index * 20 + 10), (130 + int(result[0][index] * 100), (index + 1) * 20 + 4),
(255, 0, 0), -1)
emoji_face = feelings_faces[np.argmax(result[0])]
for c in range(0, 3):
frame[200:320, 10:130, c] = emoji_face[:, :, c] * (emoji_face[:, :, 3] / 255.0) + frame[200:320, 10:130, c] * (1.0 - emoji_face[:, :, 3] / 255.0)
cv2.imshow('face', frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
def demo(modelPath, showBox=False):
face_x = tf.compat.v1.placeholder(tf.float32, [None, 2304])
y_conv = deepnn(face_x)
probs = tf.nn.softmax(y_conv)
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(modelPath)
sess = tf.Session()
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
print(
'Restore model sucsses!!\nNOTE: Press SPACE on keyboard to capture face.'
)
video_path = "test.mp4"
cap = cv2.VideoCapture(video_path)
fps = cap.get(cv2.CAP_PROP_FPS) # 获取视频的帧率
size = (int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)),
int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))) # 获取视频的大小
fourcc = cv2.VideoWriter_fourcc(*'mpeg') # 要保存的视频格式
# 把处理过的视频保存下来
output_viedo = cv2.VideoWriter()
# 保存的视频地址
video_save_path = 'trans.mp4'
output_viedo.open(video_save_path, fourcc, fps, size, True)
result = None
while True:
ret, frame = cap.read()
detected_face, face_coor = format_image(frame)
if showBox:
if face_coor is not None:
specify(face_coor, frame)
if detected_face is not None:
tensor = image_to_tensor(detected_face)
result = sess.run(probs, feed_dict={face_x: tensor})
print(result)
output_viedo.write(frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
def demo(modelPath, showBox=True):
face_x = tf.placeholder(tf.float32, [None, 2304])
y_conv = deepnn(face_x)
probs = tf.nn.softmax(y_conv)
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(modelPath)
sess = tf.Session()
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
print(
'Restore model sucsses!!\nNOTE: Press SPACE on keyboard to capture face.'
)
feelings_faces = []
for index, emotion in enumerate(EMOTIONS):
feelings_faces.append(
cv2.imread('./data/emojis/' + emotion + '.png', -1))
# video_captor = cv2.VideoCapture(0)
demo_image = cv2.imread('./demo_image.jpg')
detected_face, face_coor = format_image(demo_image)
if showBox:
if face_coor is not None:
[x, y, w, h] = face_coor
cv2.rectangle(demo_image, (x, y), (x + w, y + h), (255, 0, 0), 2)
cv2.imshow('demo_image', demo_image)
cv2.waitKey(0)
cv2.destroyAllWindows()
if detected_face is not None:
cv2.imwrite('a.jpg', detected_face)
tensor = image_to_tensor(detected_face)
result = sess.run(y_conv, feed_dict={face_x: tensor})
print('result.shape: ', result.shape)
print('result: ', result)
most_likely_index = np.argmax(result)
# print(most_likely_index)
print(EMOTIONS[most_likely_index])
sess.close()
def main(argv):
imvalue = load_data(argv)
x = tf.placeholder(tf.float32, [None, 784])
y_ = tf.placeholder(tf.float32, [None, 10])
y_conv, keep_prob = model.deepnn(x)
y_predict = tf.nn.softmax(y_conv)
init_op = tf.global_variables_initializer()
saver = tf.train.Saver()
with tf.Session() as sess:
sess.run(init_op)
saver.restore(sess, "./mnist_cnn_model.ckpt")
#saver=tf.train.import_meta_graph("E:/Project/Python/mnist/MNIST_my/mnist_cnn_model.ckpt.meta")
prediction = tf.argmax(y_predict, 1)
predint = prediction.eval(feed_dict={
x: [imvalue],
keep_prob: 1.0
},
session=sess)
print(predint[0])
def demo(modelPath, showBox=False):
face_x = tf.placeholder(tf.float32, [None, 2304]) # 定义一个TensorFlow的占位符
y_conv = deepnn(face_x) # 输出一个维度为（1*7）的结果
probs = tf.nn.softmax(y_conv) # 使用softmax激活函数将结果计算为哪种表情的概率
saver = tf.train.Saver() # 定义一个Saver对象
ckpt = tf.train.get_checkpoint_state(
modelPath) # 得到check_point路径，生成一个check_point对象
sess = tf.Session() # 定义一个TensorFlow的对话，在对话中去执行代码和训练模型
if ckpt and ckpt.model_checkpoint_path:
# 如果check_point对象存在，并且check_point的路径存在
saver.restore(
sess, ckpt.model_checkpoint_path) # 保存Session对话 这一行代码可能实在训练完之后才会执行
print(
'Restore model sucsses!!\nNOTE: Press SPACE on keyboard to capture face.'
)
feelings_faces = []
for index, emotion in enumerate(EMOTIONS):
# 将EMOTION 转换为枚举类型后提取 索引和值
feelings_faces.append(
cv2.imread('./data/emojis/' + emotion + '.png', -1))
# cv2.imread（）读取emojis文件夹下的表情图像转化为二进制数据存储到felling_facs中
video_captor = cv2.VideoCapture(0) # CV2.VideoCapture（）打开摄像头获取照片 按空格拍照
emoji_face = []
result = None
while True:
ret, frame = video_captor.read() # 读取拍到的照片转化为二进制数据 存储到frame中
detected_face, face_coor = format_image(
frame) # format_image（）函数定义在 14 行，得到脸部图像和坐标
if showBox:
if face_coor is not None:
[x, y, w, h] = face_coor
cv2.rectangle(frame, (x, y), (x + w, y + h), (255, 0, 0), 2)
if cv2.waitKey(1) & 0xFF == ord(' '): # 不知道干嘛用的，以后再说
if detected_face is not None:
cv2.imwrite('a.jpg', detected_face) # 保存脸的图像
tensor = image_to_tensor(detected_face)
# 将图像转化为一个48*48的0到255灰度图像并转化为np.array数据
result = sess.run(probs, feed_dict={
face_x: tensor
}) # 运行TensorFlow模型，计算表情的概率，返回模型训练结果
# print(result)
if result is not None:
for index, emotion in enumerate(EMOTIONS):
cv2.putText(frame, emotion, (10, index * 20 + 20),
cv2.FONT_HERSHEY_PLAIN, 0.5, (0, 255, 0), 1)
# 将人脸的表情的文字添加到人脸图片上， 参数是什么以后站再说
cv2.rectangle(frame, (130, index * 20 + 10),
(130 + int(result[0][index] * 100),
(index + 1) * 20 + 4), (255, 0, 0), -1)
emoji_face = feelings_faces[np.argmax(result[0])]
for c in range(0, 3):
frame[200:320, 10:130, c] = emoji_face[:, :, c] * (
emoji_face[:, :, 3] /
255.0) + frame[200:320, 10:130,
c] * (1.0 - emoji_face[:, :, 3] / 255.0)
cv2.imshow('face', frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
def demo(modelPath, showBox=False):
face_x = tf.placeholder(tf.float32, [None, 2304])
y_conv = deepnn(face_x)
probs = tf.nn.softmax(y_conv)
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(modelPath)
sess = tf.Session()
# Get the model
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
print('Restore model sucsses!!')
# print('NOTE: Press SPACE on keyboard to capture face.')
feelings_faces = []
for index, emotion in enumerate(EMOTIONS):
feelings_faces.append(
cv2.imread('./data/emojis/' + emotion + '.png', -1))
video_captor = cv2.VideoCapture(0)
emoji_face = []
result = None
while True:
ret, frame = video_captor.read()
# time.sleep(0.1)
detected_face, face_coor = format_image(frame)
if showBox:
if face_coor is not None:
[x, y, w, h] = face_coor
cv2.rectangle(frame, (x, y), (x + w, y + h), (255, 0, 0), 2)
# if cv2.waitKey(1) & 0xFF == ord(' '):
# Not using because it calls the model again and again
if detected_face is not None:
# cv2.imwrite('a.jpg', detected_face)
tensor = image_to_tensor(detected_face)
result = sess.run(probs, feed_dict={face_x: tensor})
# print(result)
if result is not None:
for index, emotion in enumerate(EMOTIONS):
# Label(emotion) display
cv2.putText(frame, emotion, (10, index * 20 + 20),
cv2.FONT_HERSHEY_PLAIN, 0.5, (0, 255, 0), 1)
# Histogram display
cv2.rectangle(frame, (130, index * 20 + 10),
(130 + int(result[0][index] * 100),
(index + 1) * 20 + 4), (255, 0, 0), -1)
# emoji<-emotion of the highest score
emoji_face = feelings_faces[np.argmax(result[0])]
# Emoji(120*120) display
for c in range(0, 3):
frame[200:320, 10:130, c] = emoji_face[:, :, c] * (
emoji_face[:, :, 3] /
255.0) + frame[200:320, 10:130,
c] * (1.0 - emoji_face[:, :, 3] / 255.0)
# Real-time capture
cv2.imshow('face', frame)
if cv2.waitKey(1) & 0xFF == ord(' '):
cv2.imwrite('fer_result/' + str(uuid.uuid4().hex) + '.png', frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
caffe.set_mode_gpu()
PNet = caffe.Net(caffe_model_path + "/det1.prototxt", caffe_model_path + "/det1.caffemodel", caffe.TEST)
RNet = caffe.Net(caffe_model_path + "/det2.prototxt", caffe_model_path + "/det2.caffemodel", caffe.TEST)
ONet = caffe.Net(caffe_model_path + "/det3.prototxt", caffe_model_path + "/det3.caffemodel", caffe.TEST)
owd = os.getcwd()
face_model_path = 'models/recognition/'
face_model = './face_deploy.prototxt'
face_weights = './85_accuracy.caffemodel'
center_facenet = caffe.Net(face_model_path + face_model, face_model_path + face_weights, caffe.TEST)
if EXPRESSION_DETECTION_ENABLED:
face_x = tf.placeholder(tf.float32, [None, 2304])
y_conv = deepnn(face_x)
probs = tf.nn.softmax(y_conv)
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state('models/expression/ckpt')
EMOTIONS = ['Angry', 'Disgusted', 'Fearful', 'Happy', 'Sad', 'Surprised', 'Neutral']
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.333)
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options))
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
print('Restore model sucsses!!\nNOTE: Press SPACE on keyboard to capture face.')
def drawBoxes(im, boxes):
def demo(modelPath, showBox=True):
face_x = tf.placeholder(tf.float32, [None, 2304])
y_conv = deepnn(face_x)
probs = tf.nn.softmax(y_conv)
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(modelPath)
sess = tf.Session()
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
print(
'Restore model sucsses!!\nNOTE: Press SPACE on keyboard to capture face.'
)
# feelings_faces = []
# for index, emotion in enumerate(EMOTIONS):
# num = random.randint(1,3)
# feelings_faces.append(cv2.imread('./data/emojis/' + emotion + str(num) + '.jpg', -1))
# emoji_face = []
result = None
video_captor = cv2.VideoCapture(0)
while True:
ret, frame = video_captor.read()
detected_face, face_coor = format_image(frame)
window_name = 'Face Expression Recognition'
cv2.namedWindow(window_name, cv2.WND_PROP_FULLSCREEN)
cv2.moveWindow(window_name, 0, 0)
if showBox:
if face_coor is not None:
[x, y, w, h] = face_coor
cv2.rectangle(frame, (x, y), (x + w, y + h), (255, 0, 0), 2)
ratio = 0.8
frame = cv2.resize(frame,
None,
fx=ratio,
fy=ratio,
interpolation=cv2.INTER_CUBIC)
move_dx = (1 - ratio) * 1280
move_dy = (1 - ratio) * 360
M = np.float32([[1, 0, move_dx], [0, 1, move_dy]])
new_height, new_width, _ = frame.shape
frame = cv2.warpAffine(frame, M, (1280, 720))
cv2.rectangle(frame, (0, 0), (int(move_dx), 720), (255, 255, 255), -1)
if cv2.waitKey(10) & 0xFF == ord(' '):
if detected_face is not None:
cv2.imwrite('a.jpg', detected_face)
tensor = image_to_tensor(detected_face)
result = sess.run(probs, feed_dict={face_x: tensor})
# print(result)
if result is not None:
for index, emotion in enumerate(EMOTIONS):
cv2.putText(frame, emotion, (10, index * 20 + 100),
cv2.FONT_HERSHEY_PLAIN, 1.0, (0, 0, 0), 1)
cv2.rectangle(frame, (130, index * 20 + 90),
(130 + int(result[0][index] * 100),
(index + 1) * 20 + 84), (255, 0, 0), -1)
emotion_type = EMOTIONS[np.argmax(result[0])]
prob = float(np.max(result[0]) * 100)
cv2.putText(frame, emotion_type, (20, 490),
cv2.FONT_HERSHEY_PLAIN, 1.5, (0, 0, 0), 1)
cv2.putText(frame, str('%.2f' % prob + "%"), (150, 490),
cv2.FONT_HERSHEY_PLAIN, 1.5, (0, 0, 0), 1)
# emoji_face = feelings_faces[np.argmax(result[0])]
# print("1")
# for c in range(0, 3):
# frame[200:320, 10:130, c] = emoji_face[:, :, c] * (emoji_face[:, :, 3] / 255.0) + frame[200:320, 10:130, c] * (1.0 - emoji_face[:, :, 3] / 255.0)
detected_img = cv2.imread('a.jpg', 0)
detected_img = cv2.resize(detected_img,
None,
fx=2,
fy=2,
interpolation=cv2.INTER_CUBIC)
for c in range(0, 3):
frame[340:436,
int(move_dx / 2) - 48:int(move_dx / 2) + 48,
c] = detected_img
cv2.imshow(window_name, frame)
if cv2.waitKey(10) & 0xFF == ord('q'):
break
import model
mnist = input_data.read_data_sets('MNIST_data', one_hot=True)#读取数据
#输入变量，mnist图片大小为28*28
x = tf.placeholder(tf.float32, [None, 784])
#输出变量，数字是1-10
y_ = tf.placeholder(tf.float32, [None, 10])
# 构建网络，输入—>第一层卷积—>第一层池化—>第二层卷积—>第二层池化—>第一层全连接—>第二层全连接
y_conv, keep_prob = model.deepnn(x)
#第一步对网络最后一层的输出做一个softmax，第二步将softmax输出和实际样本做一个交叉熵
#cross_entropy返回的是向量
with tf.name_scope('loss'):
cross_entropy = tf.nn.softmax_cross_entropy_with_logits(labels=y_,
logits=y_conv)
#求cross_entropy向量的平均值得到交叉熵
cross_entropy = tf.reduce_mean(cross_entropy)
#AdamOptimizer是Adam优化算法：一个寻找全局最优点的优化算法，引入二次方梯度校验
with tf.name_scope('adam_optimizer'):
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
#在测试集上的精确度
maxval=0.03,
dtype=tf.float32)
x_in = tf.add(
x_in,
tf.random_normal(shape=tf.shape(x_in),
mean=0.0,
stddev=std,
dtype=tf.float32))
x_in = tf.minimum(x_in, 1.0)
x_in = tf.maximum(x_in, 0.0)
y_ = tf.placeholder(tf.int64, [None])
# Network specification based on hyper-parameters
with tf.variable_scope("model"):
y, keep_prob = model.deepnn(x_in, fsize, ldepth, lwidth, bn, act, itype,
True, True)
for v in tf.global_variables():
print("Name: ", v.name)
print('Model size = %d weights\n' % model.count_all_vars())
sys.stdout.flush()
# Loss
with tf.name_scope('loss'):
cross_entropy = tf.losses.sparse_softmax_cross_entropy(labels=y_, logits=y)
cross_entropy = tf.reduce_mean(cross_entropy)
# Optimizer specification
with tf.variable_scope("optimizer"):
extra_update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
def demo(modelPath, showBox=False):
face_x = tf.placeholder(dtype=tf.float32, name='inputs', shape=[None, 2304])
#parser = argparse.ArgumentParser()
#parser.add_argument("--fz_model_fn",default = "./emotion_model_frozen.pb",type=str,help="Frozen model file to import")
#args = parser.parse_args()
#graph = load_graph(args.fz_model_fn)
#for op in graph.get_operations():
#print(op.name,op.values())
#x = graph.get_tensor_by_name('prefix/inputs:0')
#probs = graph.get_tensor_by_name('prefix/output_node:0')
y_conv = deepnn(face_x)
probs = tf.nn.softmax(y_conv, name='output_node')
print("probs",probs)
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(modelPath)
sess = tf.Session()
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
print('Restore model sucsses!!\nNOTE: Press SPACE on keyboard to capture face.')
feelings_faces = []
for index, emotion in enumerate(EMOTIONS):
feelings_faces.append(cv2.imread('./data/emojis/' + emotion + '.png', -1))
video_captor = cv2.VideoCapture(0)
emoji_face = []
result = None
while True:
ret, frame = video_captor.read()
detected_face, face_coor = format_image(frame)
if showBox:
if face_coor is not None:
[x,y,w,h] = face_coor
cv2.rectangle(frame, (x,y), (x+w,y+h), (255,0,0), 2)
if cv2.waitKey(1):
if detected_face is not None:
cv2.imwrite('a.jpg', detected_face)
tensor = image_to_tensor(detected_face)
result = sess.run(probs, feed_dict={face_x: tensor})
print(result)
print("probs",probs)
tf.train.write_graph(sess.graph_def, './', "nn_model.pbtxt", as_text=True)
for op in tf.get_default_graph().get_operations():
print(op.name)
# print(result)
if result is not None:
for index, emotion in enumerate(EMOTIONS):
cv2.putText(frame, emotion, (10, index * 20 + 20), cv2.FONT_HERSHEY_PLAIN, 0.5, (0, 255, 0), 1)
cv2.rectangle(frame, (130, index * 20 + 10), (130 + int(result[0][index] * 100), (index + 1) * 20 + 4),
(255, 0, 0), -1)
emoji_face = feelings_faces[np.argmax(result[0])]
for c in range(0, 3):
frame[200:320, 10:130, c] = emoji_face[:, :, c] * (emoji_face[:, :, 3] / 255.0) + frame[200:320, 10:130, c] * (1.0 - emoji_face[:, :, 3] / 255.0)
cv2.imshow('face', frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
