def prepare(self):
if FLAGS.mode == 'test':
FLAGS.batch_size = 1
self.learning_rate = tf.placeholder(tf.float32, name='learning_rate')
self.images = tf.placeholder(
tf.float32, [2, FLAGS.batch_size, IMAGE_HEIGHT, IMAGE_WIDTH, 3],
name='images')
self.labels = tf.placeholder(tf.float32, [FLAGS.batch_size, 2],
name='labels')
self.is_train = tf.placeholder(tf.bool, name='is_train')
self.global_step = tf.Variable(0, name='global_step', trainable=False)
weight_decay = 0.0005
self.tarin_num_id = 0
val_num_id = 0
if FLAGS.mode == 'train':
self.tarin_num_id = cuhk03_dataset.get_num_id(
FLAGS.data_dir, 'train')
elif FLAGS.mode == 'val':
val_num_id = cuhk03_dataset.get_num_id(FLAGS.data_dir, 'val')
images1, images2 = self.preprocess(self.images, self.is_train)
print('Build network')
logits = self.network(images1, images2, weight_decay)
self.loss = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(labels=self.labels,
logits=logits))
self.inference = tf.nn.softmax(logits)
optimizer = tf.train.MomentumOptimizer(self.learning_rate,
momentum=0.9)
self.train = optimizer.minimize(self.loss,
global_step=self.global_step)
lr = FLAGS.learning_rate
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
self.gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.4)
def main(argv=None):
if FLAGS.mode == 'test':
FLAGS.batch_size = 1
learning_rate = tf.placeholder(tf.float32, name='learning_rate')
images = tf.placeholder(tf.float32, [2, FLAGS.batch_size, IMAGE_HEIGHT, IMAGE_WIDTH, 3], name='images')
labels = tf.placeholder(tf.float32, [FLAGS.batch_size, 2], name='labels')
is_train = tf.placeholder(tf.bool, name='is_train')
global_step = tf.Variable(0, name='global_step', trainable=False)
weight_decay = 0.0005
tarin_num_id = 0
val_num_id = 0
if FLAGS.mode == 'train':
tarin_num_id = cuhk03_dataset.get_num_id(FLAGS.data_dir, 'train')
elif FLAGS.mode == 'val':
val_num_id = cuhk03_dataset.get_num_id(FLAGS.data_dir, 'val')
images1, images2 = preprocess(images, is_train)
print('Build network')
logits = network(images1, images2, weight_decay)
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=labels, logits=logits))
inference = tf.nn.softmax(logits)
optimizer = tf.train.MomentumOptimizer(learning_rate, momentum=0.9)
train = optimizer.minimize(loss, global_step=global_step)
lr = FLAGS.learning_rate
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(FLAGS.logs_dir)
if ckpt and ckpt.model_checkpoint_path:
print('Restore model')
saver.restore(sess, ckpt.model_checkpoint_path)
if FLAGS.mode == 'train':
step = sess.run(global_step)
for i in range(step, FLAGS.max_steps + 1):
batch_images, batch_labels = cuhk03_dataset.read_data(FLAGS.data_dir, 'train', tarin_num_id,
IMAGE_WIDTH, IMAGE_HEIGHT, FLAGS.batch_size)
feed_dict = {learning_rate: lr, images: batch_images,
labels: batch_labels, is_train: True}
sess.run(train, feed_dict=feed_dict)
train_loss = sess.run(loss, feed_dict=feed_dict)
print('Step: %d, Learning rate: %f, Train loss: %f' % (i, lr, train_loss))
lr = FLAGS.learning_rate * ((0.0001 * i + 1) ** -0.75)
if i % 1000 == 0:
saver.save(sess, FLAGS.logs_dir + 'model.ckpt', i)
elif FLAGS.mode == 'val':
total = 0.
for _ in range(10):
batch_images, batch_labels = cuhk03_dataset.read_data(FLAGS.data_dir, 'val', val_num_id,
IMAGE_WIDTH, IMAGE_HEIGHT, FLAGS.batch_size)
feed_dict = {images: batch_images, labels: batch_labels, is_train: False}
prediction = sess.run(inference, feed_dict=feed_dict)
prediction = np.argmax(prediction, axis=1)
label = np.argmax(batch_labels, axis=1)
for i in range(len(prediction)):
if prediction[i] == label[i]:
total += 1
print('Accuracy: %f' % (total / (FLAGS.batch_size * 10)))
elif FLAGS.mode == 'test':
image1 = cv2.imread(FLAGS.image1)
image1 = cv2.resize(image1, (IMAGE_WIDTH, IMAGE_HEIGHT))
image1 = cv2.cvtColor(image1, cv2.COLOR_BGR2RGB)
image1 = np.reshape(image1, (1, IMAGE_HEIGHT, IMAGE_WIDTH, 3)).astype(float)
image2 = cv2.imread(FLAGS.image2)
image2 = cv2.resize(image2, (IMAGE_WIDTH, IMAGE_HEIGHT))
image2 = cv2.cvtColor(image2, cv2.COLOR_BGR2RGB)
image2 = np.reshape(image2, (1, IMAGE_HEIGHT, IMAGE_WIDTH, 3)).astype(float)
test_images = np.array([image1, image2])
feed_dict = {images: test_images, is_train: False}
prediction = sess.run(inference, feed_dict=feed_dict)
print(bool(not np.argmax(prediction[0])))
def main(argv=None):
if FLAGS.mode == 'test':
FLAGS.batch_size = 1
learning_rate = tf.placeholder(tf.float32, name='learning_rate')
images = tf.placeholder(
tf.float32, [2, FLAGS.batch_size, IMAGE_HEIGHT, IMAGE_WIDTH, 3],
name='images')
labels = tf.placeholder(tf.float32, [FLAGS.batch_size, 2], name='labels')
is_train = tf.placeholder(tf.bool, name='is_train')
global_step = tf.Variable(0, name='global_step', trainable=False)
weight_decay = 0.0005
tarin_num_id = 0
val_num_id = 0
if FLAGS.mode == 'train':
tarin_num_id = cuhk03_dataset.get_num_id(FLAGS.data_dir, 'train')
elif FLAGS.mode == 'val':
val_num_id = cuhk03_dataset.get_num_id(FLAGS.data_dir, 'val')
images1, images2 = preprocess(images, is_train)
print('Build network')
logits = network(images1, images2, weight_decay)
loss = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(labels=labels, logits=logits))
inference = tf.nn.softmax(logits)
optimizer = tf.train.MomentumOptimizer(learning_rate, momentum=0.9)
train = optimizer.minimize(loss, global_step=global_step)
lr = FLAGS.learning_rate
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(FLAGS.logs_dir)
if ckpt and ckpt.model_checkpoint_path:
print('Restore model')
saver.restore(sess, ckpt.model_checkpoint_path)
if FLAGS.mode == 'train':
step = sess.run(global_step)
for i in xrange(step, FLAGS.max_steps + 1):
batch_images, batch_labels = cuhk03_dataset.read_data(
FLAGS.data_dir, 'train', tarin_num_id, IMAGE_WIDTH,
IMAGE_HEIGHT, FLAGS.batch_size)
feed_dict = {
learning_rate: lr,
images: batch_images,
labels: batch_labels,
is_train: True
}
sess.run(train, feed_dict=feed_dict)
train_loss = sess.run(loss, feed_dict=feed_dict)
print('Step: %d, Learning rate: %f, Train loss: %f' %
(i, lr, train_loss))
lr = FLAGS.learning_rate * ((0.0001 * i + 1)**-0.75)
if i % 1000 == 0:
saver.save(sess, FLAGS.logs_dir + 'model.ckpt', i)
elif FLAGS.mode == 'val':
total = 0.
for _ in xrange(10):
batch_images, batch_labels = cuhk03_dataset.read_data(
FLAGS.data_dir, 'val', val_num_id, IMAGE_WIDTH,
IMAGE_HEIGHT, FLAGS.batch_size)
feed_dict = {
images: batch_images,
labels: batch_labels,
is_train: False
}
prediction = sess.run(inference, feed_dict=feed_dict)
prediction = np.argmax(prediction, axis=1)
label = np.argmax(batch_labels, axis=1)
for i in xrange(len(prediction)):
if prediction[i] == label[i]:
total += 1
print('Accuracy: %f' % (total / (FLAGS.batch_size * 10)))
'''
for i in xrange(len(prediction)):
print('Prediction: %s, Label: %s' % (prediction[i] == 0, labels[i] == 0))
image1 = cv2.cvtColor(batch_images[0][i], cv2.COLOR_RGB2BGR)
image2 = cv2.cvtColor(batch_images[1][i], cv2.COLOR_RGB2BGR)
image = np.concatenate((image1, image2), axis=1)
cv2.imshow('image', image)
key = cv2.waitKey(0)
if key == 1048603: # ESC key
break
'''
elif FLAGS.mode == 'test':
# image_path1='20180516_155201_CH05_pic_recog/person/'
# image_path2='../persons/'#'20180516_155203_CH19_pic_recog/person/'
# save_path1='20180516_sameperson/CH05/'
# save_path2='20180718/Preview_192.168.7.27_0_20180718_220140_3056687/'#'20180516_sameperson/CH19/'
# if not os.path.exists(save_path1):
# os.makedirs(save_path1)
# if not os.path.exists(save_path2):
# os.makedirs(save_path2)
# image_files1=os.listdir(image_path1)
# image_files2=os.listdir(image_path2)
# flen1=len(image_files1)
# flen2=len(image_files2)
# print(flen1)
# print(flen2)
print '--------end---------'
image1 = cv2.imread(FLAGS.image1)
image1 = cv2.resize(image1, (IMAGE_WIDTH, IMAGE_HEIGHT))
image1 = cv2.cvtColor(image1, cv2.COLOR_BGR2RGB)
image1 = np.reshape(
image1, (1, IMAGE_HEIGHT, IMAGE_WIDTH, 3)).astype(float)
image2 = cv2.imread(FLAGS.image2)
image2 = cv2.resize(image2, (IMAGE_WIDTH, IMAGE_HEIGHT))
image2 = cv2.cvtColor(image2, cv2.COLOR_BGR2RGB)
image2 = np.reshape(
image2, (1, IMAGE_HEIGHT, IMAGE_WIDTH, 3)).astype(float)
test_images = np.array([image1, image2])
merged_summary_op = tf.summary.merge_all()
writer = tf.summary.FileWriter("./logs/", sess.graph)
# print("test_images : "+str(test_images))
#image_shaped_input = tf.reshape(test_images, [-1, IMAGE_HEIGHT, IMAGE_WIDTH, 3])
#tf.summary.image('input1', image_shaped_input, 2)
# merged = tf.summary.merge_all()
# feed_dict = {images: test_images, is_train: False}
# summary=sess.run(merged, feed_dict=feed_dict)
feed_dict = {images: test_images, is_train: False}
summary, prediction = sess.run([merged_summary_op, inference],
feed_dict=feed_dict)
print(bool(not np.argmax(prediction[0])))
writer.add_summary(summary)
writer.close()
def main(argv=None):
if FLAGS.mode == 'test':
FLAGS.batch_size = 1
learning_rate = tf.placeholder(tf.float32, name='learning_rate')
images = tf.placeholder(tf.float32, [2, FLAGS.batch_size, IMAGE_HEIGHT, IMAGE_WIDTH, 3], name='images')
labels = tf.placeholder(tf.float32, [FLAGS.batch_size, 2], name='labels')
is_train = tf.placeholder(tf.bool, name='is_train')
global_step = tf.Variable(0, name='global_step', trainable=False)
weight_decay = 0.0005
tarin_num_id = 0
val_num_id = 0
if FLAGS.mode == 'train':
tarin_num_id = cuhk03_dataset.get_num_id(FLAGS.data_dir, 'train')
elif FLAGS.mode == 'val':
val_num_id = cuhk03_dataset.get_num_id(FLAGS.data_dir, 'val')
images1, images2 = preprocess(images, is_train)
print('Build network')
logits = network(images1, images2, weight_decay)
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=labels, logits=logits))
inference = tf.nn.softmax(logits)
optimizer = tf.train.MomentumOptimizer(learning_rate, momentum=0.9)
train = optimizer.minimize(loss, global_step=global_step)
lr = FLAGS.learning_rate
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(FLAGS.logs_dir)
if ckpt and ckpt.model_checkpoint_path:
print('Restore model')
saver.restore(sess, ckpt.model_checkpoint_path)
if FLAGS.mode == 'train':
step = sess.run(global_step)
for i in xrange(step, FLAGS.max_steps + 1):
batch_images, batch_labels = cuhk03_dataset.read_data(FLAGS.data_dir, 'train', tarin_num_id,
IMAGE_WIDTH, IMAGE_HEIGHT, FLAGS.batch_size)
feed_dict = {learning_rate: lr, images: batch_images,
labels: batch_labels, is_train: True}
sess.run(train, feed_dict=feed_dict)
train_loss = sess.run(loss, feed_dict=feed_dict)
print('Step: %d, Learning rate: %f, Train loss: %f' % (i, lr, train_loss))
lr = FLAGS.learning_rate * ((0.0001 * i + 1) ** -0.75)
if i % 1000 == 0:
saver.save(sess, FLAGS.logs_dir + 'model.ckpt', i)
elif FLAGS.mode == 'val':
total = 0.
for _ in xrange(10):
batch_images, batch_labels = cuhk03_dataset.read_data(FLAGS.data_dir, 'val', val_num_id,
IMAGE_WIDTH, IMAGE_HEIGHT, FLAGS.batch_size)
feed_dict = {images: batch_images, labels: batch_labels, is_train: False}
prediction = sess.run(inference, feed_dict=feed_dict)
prediction = np.argmax(prediction, axis=1)
label = np.argmax(batch_labels, axis=1)
for i in xrange(len(prediction)):
if prediction[i] == label[i]:
total += 1
print('Accuracy: %f' % (total / (FLAGS.batch_size * 10)))
'''
for i in xrange(len(prediction)):
print('Prediction: %s, Label: %s' % (prediction[i] == 0, labels[i] == 0))
image1 = cv2.cvtColor(batch_images[0][i], cv2.COLOR_RGB2BGR)
image2 = cv2.cvtColor(batch_images[1][i], cv2.COLOR_RGB2BGR)
image = np.concatenate((image1, image2), axis=1)
cv2.imshow('image', image)
key = cv2.waitKey(0)
if key == 1048603: # ESC key
break
'''
elif FLAGS.mode == 'test':
# image_path1='20180516_155201_CH05_pic_recog/person/'
# image_path2='../persons/'#'20180516_155203_CH19_pic_recog/person/'
# save_path1='20180516_sameperson/CH05/'
# save_path2='20180718/Preview_192.168.7.27_0_20180718_220140_3056687/'#'20180516_sameperson/CH19/'
# if not os.path.exists(save_path1):
# os.makedirs(save_path1)
# if not os.path.exists(save_path2):
# os.makedirs(save_path2)
# image_files1=os.listdir(image_path1)
# image_files2=os.listdir(image_path2)
# flen1=len(image_files1)
# flen2=len(image_files2)
# print(flen1)
# print(flen2)
# flags1=[0]*(flen1)
# flags2=[0]*(flen2)
# index1_1=1#4115#1
# maxindex1=9230#4116#9230
# initpath1='20180516_155201_CH05_pic_recog/initperson/'
# init_persons1=os.listdir(initpath1)
# initp1={}
# for i in range(len(init_persons1)):
# initp1[i]=initpath1+init_persons1[i]
# new_path=save_path1+str(i)
# if not os.path.exists(new_path):
# os.makedirs(new_path)
# shutil.copyfile(initpath1+init_persons1[i],new_path+'/'+init_persons1[i])
# if not os.path.exists(save_path1+'nones'):
# os.makedirs(save_path1+'nones')
# while index1_1<maxindex1:
# maxindexes=[]
# maxscores=[]
# im2_names=[]
# for j in range(flen1):
# index1_2=int(image_files1[j].split('.jpg')[0].split('_')[-1])
# if index1_2==(index1_1+1):
# im2_names.append(image_files1[j])
# for nm2 in range(len(im2_names)):
# im2=image_path1+im2_names[nm2]
# im2_scores={}
# maxindex=-1
# maxscore=0.0
# for nm1 in range(len(initp1)):
# im1=initp1[nm1]
# y1_1=int(im1.split('_')[-6])
# x1_1=int(im1.split('_')[-5])
# y1_2=int(im1.split('_')[-4])
# x1_2=int(im1.split('_')[-3])
# y1_0=(y1_1+y1_2)/2.0
# x1_0=(x1_1+x1_2)/2.0
# y2_1=int(im2.split('_')[-6])
# x2_1=int(im2.split('_')[-5])
# y2_2=int(im2.split('_')[-4])
# x2_2=int(im2.split('_')[-3])
# y2_0=(y2_1+y2_2)/2.0
# x2_0=(x2_1+x2_2)/2.0
# dist=math.sqrt(math.pow(y1_0-y2_0,2)+math.pow(x1_0-x2_0,2))
# predtemp=0.0
# if dist>50:
# predtemp=0.0
# else:
# predtemp=cmpims(im1,im2,inference,images,is_train,sess)
# im2_scores[nm1]=predtemp
# if predtemp>maxscore and predtemp>0.9:
# maxindex=nm1
# maxscore=predtemp
# maxindexes.append(maxindex)
# maxscores.append(im2_scores)
# while True:
# moreindexes=[]
# for a in maxindexes:
# if maxindexes.count(a)>1 and (not a in moreindexes):
# moreindexes.append(a)
# if len(moreindexes)==0:
# break
# if len(moreindexes)==1 and moreindexes[0]==-1:
# break
# for a in moreindexes:
# flag=0
# list_index=[]
# for n in range(maxindexes.count(a)):
# sec=flag
# flag=maxindexes[flag:].index(a)
# list_index.append(flag+sec)
# flag=list_index[-1:][0]+1
# print str(a)+" : "+str(list_index)
# temp_index=list_index[0]
# for m in list_index:
# if(maxscores[m].get(a)>maxscores[temp_index].get(a)):
# temp_index=m
# for m in list_index:
# if m!=temp_index:
# nextindex=findNextMax(maxscores[m],maxscores[m].get(a))
# if nextindex==-1:
# maxindexes[m]=-1
# elif maxscores[m][nextindex] < 0.9:
# maxindexes[m]=-1
# else:
# maxindexes[m]=nextindex
# for m in range(len(maxindexes)):
# maxindex=maxindexes[m]
# if maxindex!=-1:
# shutil.copyfile(image_path1+im2_names[m],save_path1+str(maxindex)+'/'+im2_names[m])
# initp1[maxindex]=save_path1+str(maxindex)+'/'+im2_names[m]
# else:
# shutil.copyfile(image_path1+im2_names[m],save_path1+'nones/'+im2_names[m])
# index1_1=index1_1+1
print '--------end---------'
image1 = cv2.imread(FLAGS.image1)
image1 = cv2.resize(image1, (IMAGE_WIDTH, IMAGE_HEIGHT))
image1 = cv2.cvtColor(image1, cv2.COLOR_BGR2RGB)
image1 = np.reshape(image1, (1, IMAGE_HEIGHT, IMAGE_WIDTH, 3)).astype(float)
image2 = cv2.imread(FLAGS.image2)
image2 = cv2.resize(image2, (IMAGE_WIDTH, IMAGE_HEIGHT))
image2 = cv2.cvtColor(image2, cv2.COLOR_BGR2RGB)
image2 = np.reshape(image2, (1, IMAGE_HEIGHT, IMAGE_WIDTH, 3)).astype(float)
test_images = np.array([image1, image2])
feed_dict = {images: test_images, is_train: False}
prediction,pool1_2 = sess.run(inference, feed_dict=feed_dict)
print(bool(not np.argmax(prediction[0])))
def main(argv=None):
if FLAGS.mode == 'test':
FLAGS.batch_size = 1
if FLAGS.mode == 'data':
FLAGS.batch_size = 1
learning_rate = tf.placeholder(tf.float32, name='learning_rate')
images = tf.placeholder(tf.float32, [2, FLAGS.batch_size, IMAGE_HEIGHT, IMAGE_WIDTH, 3], name='images')
labels = tf.placeholder(tf.float32, [FLAGS.batch_size, 2], name='labels')
is_train = tf.placeholder(tf.bool, name='is_train')
global_step = tf.Variable(0, name='global_step', trainable=False)
weight_decay = 0.0005
tarin_num_id = 0
val_num_id = 0
if FLAGS.mode == 'train':
tarin_num_id = cuhk03_dataset.get_num_id(FLAGS.data_dir, 'train')
elif FLAGS.mode == 'val':
val_num_id = cuhk03_dataset.get_num_id(FLAGS.data_dir, 'val')
images1, images2 = preprocess(images, is_train)
print('=======================Build Network=======================')
logits = network(images1, images2, weight_decay)
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=labels, logits=logits))
inference = tf.nn.softmax(logits)
optimizer = tf.train.MomentumOptimizer(learning_rate, momentum=0.9)
train = optimizer.minimize(loss, global_step=global_step)
lr = FLAGS.learning_rate
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(FLAGS.logs_dir)
if ckpt and ckpt.model_checkpoint_path:
print('==================================Restore model==================================')
saver.restore(sess, ckpt.model_checkpoint_path)
if FLAGS.mode == 'train':
step = sess.run(global_step)
for i in range(step, FLAGS.max_steps + 1):
batch_images, batch_labels = cuhk03_dataset.read_data(FLAGS.data_dir, 'train', tarin_num_id,
IMAGE_WIDTH, IMAGE_HEIGHT, FLAGS.batch_size)
feed_dict = {learning_rate: lr, images: batch_images,
labels: batch_labels, is_train: True}
sess.run(train, feed_dict=feed_dict)
train_loss = sess.run(loss, feed_dict=feed_dict)
print('Step: %d, Learning rate: %f, Train loss: %f' % (i, lr, train_loss))
lr = FLAGS.learning_rate * ((0.0001 * i + 1) ** -0.75)
if i % 1000 == 0:
saver.save(sess, FLAGS.logs_dir + 'model.ckpt', i)
elif FLAGS.mode == 'val':
total = 0.
for _ in range(10):
batch_images, batch_labels = cuhk03_dataset.read_data(FLAGS.data_dir, 'val', val_num_id,
IMAGE_WIDTH, IMAGE_HEIGHT, FLAGS.batch_size)
feed_dict = {images: batch_images, labels: batch_labels, is_train: False}
prediction = sess.run(inference, feed_dict=feed_dict)
prediction = np.argmax(prediction, axis=1)
label = np.argmax(batch_labels, axis=1)
for i in range(len(prediction)):
if prediction[i] == label[i]:
total += 1
print('Accuracy: %f' % (total / (FLAGS.batch_size * 10)))
'''
for i in range(len(prediction)):
print('Prediction: %s, Label: %s' % (prediction[i] == 0, labels[i] == 0))
image1 = cv2.cvtColor(batch_images[0][i], cv2.COLOR_RGB2BGR)
image2 = cv2.cvtColor(batch_images[1][i], cv2.COLOR_RGB2BGR)
image = np.concatenate((image1, image2), axis=1)
cv2.imshow('image', image)
key = cv2.waitKey(0)
if key == 1048603: # ESC key
break
'''
elif FLAGS.mode == 'test':
image1 = cv2.imread(FLAGS.image1)
image1 = cv2.resize(image1, (IMAGE_WIDTH, IMAGE_HEIGHT))
image1 = cv2.cvtColor(image1, cv2.COLOR_BGR2RGB)
image2 = cv2.imread(FLAGS.image2)
image2 = cv2.resize(image2, (IMAGE_WIDTH, IMAGE_HEIGHT))
image2 = cv2.cvtColor(image2, cv2.COLOR_BGR2RGB)
f = plt.figure()
f.add_subplot(1,2, 1)
plt.imshow(image1)
f.add_subplot(1,2, 2)
plt.imshow(image2)
plt.show()
print("===============================Show Images==================================================")
start = time.time()
image1 = np.reshape(image1, (1, IMAGE_HEIGHT, IMAGE_WIDTH, 3)).astype(float)
image2 = np.reshape(image2, (1, IMAGE_HEIGHT, IMAGE_WIDTH, 3)).astype(float)
test_images = np.array([image1, image2])
test_images2 = np.array([image2, image1])
feed_dict = {images: test_images, is_train: False}
feed_dict2 = {images: test_images2, is_train: False}
#print(feed_dict)
prediction = sess.run(inference, feed_dict=feed_dict)
prediction2 = sess.run(inference, feed_dict=feed_dict2)
print("=======================Prediction1=======================")
print(prediction)
print(bool(not np.argmax(prediction[0])))
#print(prediction[0])
print("=======================Prediction2=======================")
print(prediction2)
print(bool(not np.argmax(prediction2[0])))
end = time.time()
print("Time in seconds: ")
print(end - start)
elif FLAGS.mode == 'data':
print("path_test:",FLAGS.path_test)
files = sorted(glob.glob('/home/oliver/Documentos/person-reid/video3_4/*.png'))
print(len(files))
image1 = cv2.imread(FLAGS.image1)
image1 = cv2.resize(image1, (IMAGE_WIDTH, IMAGE_HEIGHT))
image1 = cv2.cvtColor(image1, cv2.COLOR_BGR2RGB)
plt.imshow(image1)
plt.show()
'''
image2 = cv2.imread(FLAGS.image2)
image2 = cv2.resize(image2, (IMAGE_WIDTH, IMAGE_HEIGHT))
image2 = cv2.cvtColor(image2, cv2.COLOR_BGR2RGB)
f = plt.figure()
f.add_subplot(1,2, 1)
plt.imshow(image1)
f.add_subplot(1,2, 2)
plt.imshow(image2)
plt.show()
print("===============================Show Images==================================================")
'''
start = time.time()
image1 = np.reshape(image1, (1, IMAGE_HEIGHT, IMAGE_WIDTH, 3)).astype(float)
#image2 = np.reshape(image2, (1, IMAGE_HEIGHT, IMAGE_WIDTH, 3)).astype(float)
#list_pred=[]
#list_bool=[]
list_all = []
for x in files:
image2 = cv2.imread(x)
image2 = cv2.resize(image2, (IMAGE_WIDTH, IMAGE_HEIGHT))
image2 = cv2.cvtColor(image2, cv2.COLOR_BGR2RGB)
image2 = np.reshape(image2, (1, IMAGE_HEIGHT, IMAGE_WIDTH, 3)).astype(float)
test_images = np.array([image1, image2])
feed_dict = {images: test_images, is_train: False}
prediction = sess.run(inference, feed_dict=feed_dict)
#print(bool(not np.argmax(prediction[0])))
#list_bool.append(bool(not np.argmax(prediction[0])))
#list_pred.append(prediction[0])
if bool(not np.argmax(prediction[0])):
tupl = (x, prediction[0][0], prediction[0][1])
list_all.append(tupl)
list_all.sort(key = sortSecond , reverse = True)
end = time.time()
print("Time in seconds: ")
print(end - start)
#print (list_all)
print ("size list: ", len(list_all))
####3
#cv2.namedWindow('Person-ReID', cv2.WINDOW_FULLSCREEN)
#cv2.resizeWindow('Person-ReID', 480, 320)
####
i = 0
list_reid = []
for e in list_all:
temp_img = cv2.imread(e[0])
temp_img = cv2.cvtColor(temp_img, cv2.COLOR_BGR2RGB)
fpath, fname = os.path.split(e[0])
if (i > 15 ):
break
#plt.imshow(temp_img)
#plt.show()
#cv2.namedWindow('Person-ReID', cv2.WINDOW_NORMAL)
#cv2.imshow('Person-ReID', temp_img)
cv2.imwrite("output_query/"+fname, temp_img)
#cv2.waitKey(1)
path_f, name_f = os.path.split(e[0])
splits_coords = name_f.rsplit('_')
#print("coord: ",splits_coords)
list_reid.append(( int(splits_coords[1]), splits_coords[2], splits_coords[3], splits_coords[4], splits_coords[5]))
i = i +1
print (i, e[0]," - ", e[1], " - ", e[2])
list_reid.sort(key = sortFirst)
## sort the coords for num of frame
print (list_reid)
f_frames = sorted(glob.glob('/home/oliver/Documentos/person-reid/frames/video3/*.png'))
j = 0
cv2.namedWindow('Person-ReID', cv2.WINDOW_NORMAL)
cv2.resizeWindow('Person-ReID', 640, 480)
flag_draw = False
k = 0
###PINTO EN LOS FRAMES
for frame in f_frames:
imgFrame = cv2.imread(frame , cv2.IMREAD_UNCHANGED)
frame_p, frame_n = os.path.split(frame)
temp_f = frame_n.rsplit('.')
#cv2.imshow('Person-ReID', imgFrame)
#cv2.waitKey(1)
#print(int(temp_f[0]))
if(j < len(list_reid)):
if (int(temp_f[0]) == list_reid[j][0]):
#pintar como TRUE
print (int(temp_f[0]) ,"--entro--",j, " ", list_reid[j])
#cv2.polylines(imgFrame , [np.int0([list_reid[j][1], list_reid[j][2], list_reid[j][3], list_reid[j][4]]).reshape((-1, 1, 2))], True, (0, 255, 0), 3)
#cv2.rectangle(imgFrame,(int(list_reid[j][4]), int(list_reid[j][3])) , (int(list_reid[j][2]),int(list_reid[j][1])), (0, 255, 0), 3)
#cv2.rectangle(imgFrame,(int(list_reid[j][3]), int(list_reid[j][4])),(int(list_reid[j][1]), int(list_reid[j][2])), (0, 255, 0), 3)
#color = cv2.cvtColor(np.uint8([[[num_random, 128, 200]]]),cv2.COLOR_HSV2RGB).squeeze().tolist()
#####################
#color = cv2.cvtColor(np.uint8([[[0, 128, 200]]]),cv2.COLOR_HSV2RGB).squeeze().tolist()
cv2.rectangle(imgFrame, (int(list_reid[j][3]), int(list_reid[j][1])) , (int(list_reid[j][4]),int(list_reid[j][2])) , (0,255,0), 10)
#cv2.imwrite('outReid/'+temp_f[0]+'.png',imgFrame)
flag_draw = True
k = 0
j=j+1
#else:
#cv2.imwrite('outReid/'+temp_f[0]+'.png',imgFrame)
# cv2.imshow('Person-ReID', imgFrame)
# cv2.waitKey(1)
#else:
#cv2.imshow('Person-ReID', imgFrame)
#cv2.waitKey(1)
#cv2.imwrite('outReid/'+temp_f[0]+'.png',imgFrame)
if (flag_draw == True):
font = cv2.FONT_HERSHEY_SIMPLEX
cv2.putText(imgFrame,'True',(200,200), font, 4,(0,255,0),4,cv2.LINE_AA)
k = k + 1
if (k > 15):
flag_draw = False
k = 0
#cv2.imwrite('outReid/'+temp_f[0]+'.png',imgFrame)
cv2.imshow('Person-ReID', imgFrame)
cv2.waitKey(1)
#print(e[0]," , ", e[1], "\n")
#i=0
#for x in list_bool:
# if x==True:
# print(files[i],list_pred[i],list_bool[i])
# i=i+1
#test_images = np.array([image1, image2])
#test_images2 = np.array([image2, image1])
#feed_dict = {images: test_images, is_train: False}
#feed_dict2 = {images: test_images2, is_train: False}
#print(feed_dict)
#prediction = sess.run(inference, feed_dict=feed_dict)
#prediction2 = sess.run(inference, feed_dict=feed_dict2)
print("=======================Prediction List=======================")
def main():
if FLAGS.mode == 'test':
FLAGS.batch_size = 1
learning_rate = tf.placeholder(tf.float32, name='learning_rate')
images = tf.placeholder(
tf.float32, [2, FLAGS.batch_size, IMAGE_HEIGHT, IMAGE_WIDTH, 3],
name='images')
labels = tf.placeholder(tf.float32, [FLAGS.batch_size, 2], name='labels')
is_train = tf.placeholder(tf.bool, name='is_train')
global_step = tf.Variable(0, name='global_step', trainable=False)
weight_decay = 0.0005
tarin_num_id = 0
val_num_id = 0
if FLAGS.mode == 'train':
tarin_num_id = cuhk03_dataset.get_num_id(FLAGS.data_dir, 'train')
elif FLAGS.mode == 'val':
val_num_id = cuhk03_dataset.get_num_id(FLAGS.data_dir, 'val')
images1, images2 = preprocess(images, is_train)
print('Build network')
logits = network(images1, images2, weight_decay)
loss = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(labels=labels, logits=logits))
inference = tf.nn.softmax(logits)
optimizer = tf.train.MomentumOptimizer(learning_rate, momentum=0.9)
train = optimizer.minimize(loss, global_step=global_step)
lr = FLAGS.learning_rate
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.4)
with tf.Session(config=tf.ConfigProto(gpu_options=gpu_options)) as sess:
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(FLAGS.logs_dir)
if ckpt and ckpt.model_checkpoint_path:
print('Restore model')
saver.restore(sess, ckpt.model_checkpoint_path)
if FLAGS.mode == 'train':
step = sess.run(global_step)
for i in xrange(step, FLAGS.max_steps + 1):
batch_images, batch_labels = cuhk03_dataset.read_data(
FLAGS.data_dir, 'train', tarin_num_id, IMAGE_WIDTH,
IMAGE_HEIGHT, FLAGS.batch_size)
feed_dict = {
learning_rate: lr,
images: batch_images,
labels: batch_labels,
is_train: True
}
sess.run(train, feed_dict=feed_dict)
train_loss = sess.run(loss, feed_dict=feed_dict)
print('Step: %d, Learning rate: %f, Train loss: %f' %
(i, lr, train_loss))
lr = FLAGS.learning_rate * ((0.0001 * i + 1)**-0.75)
if i % 1000 == 0:
saver.save(sess, FLAGS.logs_dir + 'model.ckpt', i)
elif FLAGS.mode == 'val':
total = 0.
for _ in xrange(10):
batch_images, batch_labels = cuhk03_dataset.read_data(
FLAGS.data_dir, 'val', val_num_id, IMAGE_WIDTH,
IMAGE_HEIGHT, FLAGS.batch_size)
feed_dict = {
images: batch_images,
labels: batch_labels,
is_train: False
}
prediction = sess.run(inference, feed_dict=feed_dict)
prediction = np.argmax(prediction, axis=1)
label = np.argmax(batch_labels, axis=1)
for i in xrange(len(prediction)):
if prediction[i] == label[i]:
total += 1
print('Accuracy: %f' % (total / (FLAGS.batch_size * 10)))
'''
for i in xrange(len(prediction)):
print('Prediction: %s, Label: %s' % (prediction[i] == 0, labels[i] == 0))
image1 = cv2.cvtColor(batch_images[0][i], cv2.COLOR_RGB2BGR)
image2 = cv2.cvtColor(batch_images[1][i], cv2.COLOR_RGB2BGR)
image = np.concatenate((image1, image2), axis=1)
cv2.imshow('image', image)
key = cv2.waitKey(0)
if key == 1048603: # ESC key
break
'''
elif FLAGS.mode == 'test':
images_prob = {}
test_query = file_name("/workspace/zyf/person_data/test_query")
test_reference = file_name(
"/workspace/zyf/person_data/test_reference")
image1 = cv2.imread(test_query[1])
image1 = cv2.resize(image1, (IMAGE_WIDTH, IMAGE_HEIGHT))
image1 = cv2.cvtColor(image1, cv2.COLOR_BGR2RGB)
image1 = np.reshape(
image1, (1, IMAGE_HEIGHT, IMAGE_WIDTH, 3)).astype(float)
for i in range(len(test_reference)):
image_reference = cv2.imread(test_reference[i])
image_reference = cv2.resize(image_reference,
(IMAGE_WIDTH, IMAGE_HEIGHT))
image_reference = cv2.cvtColor(image_reference,
cv2.COLOR_BGR2RGB)
image_reference = np.reshape(
image_reference,
(1, IMAGE_HEIGHT, IMAGE_WIDTH, 3)).astype(float)
test_images = np.array([image1, image_reference])
feed_dict = {images: test_images, is_train: False}
prediction = sess.run(inference, feed_dict=feed_dict)
images_prob.setdefault(test_reference[i], (prediction[0])[0])
images_sorted = sorted(images_prob.iteritems(),
key=lambda asd: asd[1],
reverse=True)
print(images_sorted[0:15])
print(test_query[1])
# if __name__ == '__main__':
# tf.app.run()
