def get_fc8(img_path):
img_ready = utils.load_image(img_path)# 读入待判图，并将其处理成[None,224,224,3]的形状
with tf.Session() as sess:
images = tf.placeholder(tf.float32, [1,224,224,3])
# 这里images的占位形状为[一次喂入的图片张数，图片长，图片宽，图片通道信息红绿蓝]
vgg = vgg16.Vgg16()# 实例化vgg，运行了Vgg16类的初始化函数__ini__
vgg.forward(images)
probability = sess.run(vgg.fc8, feed_dict={images: img_ready})
return probability
def taka_generator_fig_text_batch(text_seq_list,
all_fig_path_list,
label_list,
batch_size,
fig_resize_shape,
nolabel=False):
fig_mean = np.load(
os.path.join(os.path.dirname(__file__), '..', 'output',
'train_figmean.npy'))
length = len(text_seq_list)
index = 0 # index逐渐递增，但是值每次都对length取模
while True:
index_list = []
for i in range(batch_size):
index_list.append(index + i)
#label
label = np.array(list(label_list.iloc[index_list]), dtype='int32')
label = label.reshape((batch_size, 1))
#构造文本的输入array
text_seq = text_seq_list[index_list]
text_arr = np.array(list(text_seq), dtype='int32')
# 构造图像的输入array
fig_path_list_batch = all_fig_path_list.iloc[index_list]
user_fig_arr = np.zeros(shape=(batch_size, 10, fig_resize_shape,
fig_resize_shape, 3))
for batch_index in range(len(fig_path_list_batch)):
fig_path_list = fig_path_list_batch.iloc[batch_index]
for i in range(len(fig_path_list)):
path = fig_path_list[i]
try:
fig_arr = load_image(path)
if fig_arr.shape != (fig_resize_shape, fig_resize_shape,
3):
print('?:', path)
else:
fig_arr = fig_arr - fig_mean ##################### fig归一化
user_fig_arr[batch_index, i] = fig_arr
except BaseException as e:
pass
index = (index + batch_size) % length
if nolabel:
yield [text_arr, user_fig_arr]
else:
yield ([text_arr, user_fig_arr], label)
def figs_reshaped_arr(fig_path):# 传入的是某个用户的所有图片的路径
if len(fig_path)==0:# 如果某个用户拥有的图片数为0
return np.nan
reshape_list = []
for path in fig_path:
try:
img_reshaped = utils.load_image(path)
reshape_list.append(img_reshaped)
except OSError as e:
print('img cannot open:',path,e)
pass
if len(reshape_list)==0:
return np.nan
arr = np.zeros(shape=[len(reshape_list), 224, 224, 3])
for i,reshaped in enumerate(reshape_list):
arr[i,:,:,:] = reshaped
return arr
def generator_fig_hue_hist_batch(all_fig_path_list,
label_list,
hist_count,
batch_size,
fig_resize_shape,
nolabel=False):
length = len(all_fig_path_list)
index = 0
while True:
index_list = []
for i in range(batch_size):
index_list.append(index + i)
#label
label = np.array(list(label_list.iloc[index_list]), dtype='int32')
label = label.reshape((batch_size, 1))
# 构造图像的输入array
fig_path_list_batch = all_fig_path_list.iloc[index_list]
user_fig_arr = np.zeros(shape=(batch_size, 10, hist_count))
for batch_index in range(len(fig_path_list_batch)):
fig_path_list = fig_path_list_batch.iloc[batch_index]
for i in range(len(fig_path_list)):
path = fig_path_list[i]
try:
fig_arr = load_image(path)
if fig_arr.shape != (fig_resize_shape, fig_resize_shape,
3):
print('?:', path)
else:
fig_hsv = rgb2hsv(fig_arr)
fig_hue = fig_hsv[:, :, 0]
hist_arr = list2hist(fig_hue.ravel(), hist_count)
user_fig_arr[batch_index, i] = hist_arr
except BaseException as e:
pass
index = (index + batch_size) % length
if nolabel:
yield user_fig_arr
else:
yield (user_fig_arr, label)
def pred_class(img_path):
img_ready = utils.load_image(img_path)# 读入待判图，并将其处理成[1,224,224,3]的形状
# fig = plt.figure(u"Top-5 prediction")# 准备一张图，把运行的结果可视化。
with tf.Session() as sess:
images = tf.placeholder(tf.float32, [1,224,224,3])
# 这里images的占位形状为[一次喂入的图片张数，图片长，图片宽，图片通道信息红绿蓝]
vgg = vgg16.Vgg16()# 实例化vgg，运行了Vgg16类的初始化函数__ini__
vgg.forward(images)
probability = sess.run(vgg.prob, feed_dict={images: img_ready})
# probability存着1000个概率，对应1000中每个类别的概率，索引key是labels中的健
# 用probability[0][i]遍历每个概率值
top5 = np.argsort(probability[0])[-1:-6:-1]
# np.argsort表示对列表从小到大排序，返回索引值。这里的top5存着probability列表中5个最高概率的索引
print("top:", top5)
values = []# 用来存probability中元素的值，存概率
bar_label = []# 用来存标签字典中对应的值，存名称
for n, i in enumerate(top5):
# print("n:", n)
# print("i:", i)
values.append(probability[0][i])
bar_label.append(labels[i])
print(i, ":", labels[i], "----", utils.percent(probability[0][i]))
'''
ax = fig.add_subplot(111)
# fig.add_subplot(数 数 数)，111表示：包含1行，包含1列，当前是第1个
ax.bar(range(len(values)), values, tick_label=bar_label, width=0.5, fc='g')
# ax.bar(bar的个数，bar的值，每个bar的名字，bar宽，bar色)
ax.set_ylabel(u'probabilityit')
# ax.set_ylabel(" ")设置第一列的y轴名字
ax.set_title(u'Top-5')
# ax.set_title(" ")设置子图名字
for a, b in zip(range(len(values)), values):
ax.text(a, b+0.0005, utils.percent(b), ha='center', va='bottom', fontsize=7)
# ax.text(文字x坐标，文字y坐标，文字内容，ha='center'水平方向位置，va='bottom'垂直方向位置，字号)
plt.show()
'''
return values[0], bar_label[0]
def generator_fig_text_senti_multioutput_batch(text_seq_list,
all_fig_path_list,
label_list,
batch_size,
fig_resize_shape,
nolabel=False):
fig_mean = np.load(
os.path.join(os.path.dirname(__file__), '..', 'output',
'train_figmean.npy'))
dic = json.load(open(SENTI_DIC_PATH))
length = len(text_seq_list)
index = 0 # index逐渐递增，但是值每次都对length取模
while True:
index_list = []
for i in range(batch_size):
index_list.append(index + i)
#label
label = np.array(list(label_list.iloc[index_list]), dtype='int32')
label = label.reshape((batch_size, 1))
#构造文本的输入array
text_seq = text_seq_list.iloc[index_list]
text_arr = np.array(list(text_seq), dtype='int32')
text_senti = np.zeros((text_arr.shape[0], text_arr.shape[1],
text_arr.shape[2])) # (batchsize, 100, 50)
for k in range(len(text_arr)):
sample = text_arr[k]
for i in range(sample.shape[0]):
for j in range(sample.shape[1]):
num = sample[i][j]
if dic.__contains__(num):
print('contains', num)
text_senti[k][i][j] = dic[num]
text_senti = text_senti.reshape(
(text_arr.shape[0], text_arr.shape[1], text_arr.shape[2], 1))
# 构造图像的输入array
fig_path_list_batch = all_fig_path_list.iloc[index_list]
user_fig_arr = np.zeros(shape=(batch_size, 10, fig_resize_shape,
fig_resize_shape, 3))
for batch_index in range(len(fig_path_list_batch)):
fig_path_list = fig_path_list_batch.iloc[batch_index]
for i in range(len(fig_path_list)):
path = fig_path_list[i]
try:
fig_arr = load_image(path)
if fig_arr.shape != (fig_resize_shape, fig_resize_shape,
3):
print('?:', path)
else:
fig_arr = fig_arr - fig_mean ##################### fig归一化
user_fig_arr[batch_index, i] = fig_arr
except BaseException as e:
pass
index = (index + batch_size) % length
if nolabel:
yield [text_arr, text_senti, user_fig_arr]
else:
yield ([text_arr, text_senti, user_fig_arr], [label, label, label])
