def read_copy():
if not os.path.exists(os.path.join(os.path.pardir, 'OUTPUT')):
os.mkdir(os.path.join(os.path.pardir, 'OUTPUT'))
for i in range(10): # 检查输出文件夹是否存在
if not os.path.exists(os.path.join(os.path.pardir, 'OUTPUT', str(i))):
os.mkdir(os.path.join(os.path.pardir, 'OUTPUT', str(i)))
dataset = tf.data.TFRecordDataset(RECORD_NAME)
dataset = dataset.map(read_decode)
dataset = dataset.repeat(1)
dataset = dataset.batch(5)
iterator = dataset.make_one_shot_iterator()
batch_image, batch_name, batch_label = iterator.get_next()
score_label = model(batch_image, False)
pred_label = tf.argmax(score_label, 1)
softmax = tf.nn.softmax(score_label)
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
saver = tf.train.Saver()
y_true, y_pred, names, label_prob, label_score = [], [], [], [], []
with tf.Session() as sess:
sess.run(init_op)
saver.restore(sess, os.path.join(os.path.pardir, 'model',
'model1.ckpt'))
while True:
try:
label_pred, name, label_true, pred_prob, pred_score = sess.run(
[
pred_label, batch_name, batch_label, softmax,
score_label
])
# name = str(name[0],encoding='utf8')
print(label_pred, label_true)
y_true.extend(label_true)
y_pred.extend(label_pred)
names.extend(name)
label_prob.extend(pred_prob)
label_score.extend(pred_score)
# n = os.path.split(name)
# shutil.copyfile(name,os.path.join(os.path.pardir,'OUTPUT',str(label), '%s_%s'%(n[0][3:], n[1])))
except:
print('Finish')
break
return y_true, y_pred, names, label_prob, label_score
def read_copy(folder_type):
if not os.path.exists(PATH_OUT):
os.mkdir(PATH_OUT)
if not os.path.exists(os.path.join(PATH_OUT, folder_type)):
os.mkdir(os.path.join(PATH_OUT, folder_type))
for i in range(9): # 检查输出文件夹是否存在
if not os.path.exists(os.path.join(PATH_OUT, folder_type, str(i))):
os.mkdir(os.path.join(PATH_OUT, folder_type, str(i)))
dataset = tf.data.TFRecordDataset(RECORD_NAME)
dataset = dataset.map(read_decode)
dataset = dataset.repeat(1)
dataset = dataset.batch(5)
iterator = dataset.make_one_shot_iterator()
batch_image, batch_name = iterator.get_next()
batch_score = model(batch_image, False)
batch_pred = tf.argmax(batch_score, 1)
batch_softmax = tf.nn.softmax(batch_score)
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
saver = tf.train.Saver()
names, scores, preds, softmaxs = [], [], [], []
with tf.Session() as sess:
sess.run(init_op)
saver.restore(sess, os.path.join(os.path.pardir, 'model',
'model1.ckpt'))
while True:
try:
name, pred, score, softmax = sess.run(
[batch_name, batch_pred, batch_score, batch_softmax])
name = list(map(lambda x: str(x, encoding='utf8'), name))
print(pred)
for p, n in zip(pred, name):
# n = os.path.split(name)
shutil.copyfile(
n,
os.path.join(PATH_OUT, folder_type, str(p),
os.path.split(n)[1]))
except Exception as e:
print('Finish')
# print(e)
break
def read_copy():
if not os.path.exists(os.path.join(PATH_OUT)):
os.mkdir(os.path.join(PATH_OUT))
for i in range(10): # 检查输出文件夹是否存在
if not os.path.exists(os.path.join(PATH_OUT, str(i))):
os.mkdir(os.path.join(PATH_OUT, str(i)))
dataset = tf.data.TFRecordDataset(RECORD_NAME)
dataset = dataset.map(read_decode)
dataset = dataset.repeat(1)
dataset = dataset.batch(5)
iterator = dataset.make_one_shot_iterator()
batch_image, batch_name = iterator.get_next()
score_label = model(batch_image, False)
pred_label = tf.argmax(score_label, 1)
softmax = tf.nn.softmax(score_label)
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
saver = tf.train.Saver()
y_true, y_pred, names, label_prob, label_score = [], [], [], [], []
with tf.Session() as sess:
sess.run(init_op)
saver.restore(sess, os.path.join(PATH_MODEL, 'model1.ckpt'))
while True:
try:
label_pred, name = sess.run([pred_label, batch_name])
name = list(map(decode_name, name))
# print(label_pred, name)
for p, n in zip(label_pred, name):
print(p, n)
out_n = os.path.join(
PATH_OUT, str(p), '%s_%s' %
(os.path.split(n)[0][-29:], os.path.split(n)[1]))
shutil.copyfile(n, out_n)
except:
print('Finish')
break
def read_copy():
dataset = tf.data.TFRecordDataset(RECORD_NAME)
dataset = dataset.map(read_decode)
dataset = dataset.repeat(1)
dataset = dataset.batch(5)
iterator = dataset.make_one_shot_iterator()
batch_image, batch_name = iterator.get_next()
score_label = model(batch_image, False)
pred_label = tf.argmax(score_label, 1)
softmax = tf.nn.softmax(score_label)
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
saver = tf.train.Saver()
y_true, y_pred, names, label_prob, label_score = [], [], [], [], []
with tf.Session() as sess:
sess.run(init_op)
saver.restore(sess, os.path.join(PATH_MODEL, 'model1.ckpt'))
while True:
try:
label_pred, name, label_prob = sess.run(
[pred_label, batch_name, softmax])
name = list(map(decode_name, name))
# print(label_pred, name)
label_prob = np.array(label_prob)
print(label_prob.shape)
for p, n, s in zip(label_pred, name, label_prob.max(axis=1)):
if s < 0.95: # 如果准确率低于阈值，则移除掉该图片
print(p, n, s)
out_n = os.path.join(PATH_OUT, 'other',
os.path.split(n)[1])
shutil.copyfile(n, out_n) # 复制到其他文件夹
os.remove(n) # 删除图片
except Exception as e:
print('Finish')
# print(e)
break
def read_copy(folder,IDX_5, IDX_7):
dataset = tf.data.TFRecordDataset(RECORD_NAME)
dataset = dataset.map(read_decode)
dataset = dataset.repeat(1)
dataset = dataset.batch(5)
iterator = dataset.make_one_shot_iterator()
batch_image, batch_name = iterator.get_next()
score_label = model(batch_image, False)
pred_label = tf.argmax(score_label, 1)
softmax = tf.nn.softmax(score_label)
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
saver = tf.train.Saver()
y_true, y_pred, names, label_prob, label_score = [], [], [], [], []
with tf.Session() as sess:
sess.run(init_op)
saver.restore(sess, os.path.join(PATH_MODEL,'model3.ckpt'))
while True:
try:
label_pred,name,label_prob = sess.run([pred_label, batch_name,softmax])
name = list(map(decode_name, name))
# print(label_pred, name)
label_prob = np.array(label_prob)
# print(label_prob.shape)
for p,n,s in zip(label_pred, name,label_prob.max(axis=1)):
# if s < 0.95:
print(p,n,s)
if p == 0:
img = PImage.open(n)
img = img.transpose(PImage.ROTATE_270)
if p == 2:
img = PImage.open(n)
img = img.transpose(PImage.ROTATE_90)
if folder == '5':
h = img.size[0]
w = img.size[1]
img = img.crop((int(w/6), int(h/3), int(w*0.5), int(h*0.78))).resize((500, 500))
out_n = os.path.join(PATH_OUT,folder + '_rotate',"%06d.jpg" % (IDX_5+300))
IDX_5+=1
if folder == '7':
w = img.size[0]
h = img.size[1]
img = img.crop((int(w*0.5), int(h*0.3), int(w*0.95), int(h*0.8)))
img = img.resize((500, 500))
out_n = os.path.join(PATH_OUT,folder + '_rotate',"%06d.jpg" % (IDX_7+10000))
IDX_7 +=1
img.save(out_n)
# n = os.path.split(name)
# out_n = os.path.join(PATH_OUT,'other',os.path.split(n)[1])
# shutil.copyfile(n,out_n)
# os.remove(n)
except Exception as e:
print('Finish')
print(e)
break
return IDX_5, IDX_7