def get_range():
sr = StrictRedis()
res_photos = sr.lrange('final_photos', 0, -1)
res_map = sr.lrange('map', 0, -1)
for p, m in zip(res_photos, res_map):
path = m.decode('utf-8')
key = os.path.split(path)[0]
sr.rpush(key, p)
sr.rpush('{}_index'.format(key), path)
logger.info('处理完成》》》{}'.format(path))
def readPhotos(dir_path='h:/photo', count=-1):
if os.path.exists(dir_path):
droot = os.listdir(dir_path)
count = count if count != -1 else len(droot)
for d_name in droot:
d_path = os.path.join(dir_path, d_name)
d = os.listdir(d_path)
for file_name in d[:count]:
file = os.path.join(d_path, file_name)
yield d_name, file
else:
logger.info(f'目录{dir_path}:不存在')
def deal_again(start=0, end=-1):
sr = StrictRedis()
if start == 0:
sr.delete('final_photos')
res = np.array(
list(
np.fromstring(i, dtype=np.float64)
for i in sr.lrange('deal_photos', start, end)))
logger.info('训练矩阵:{}'.format(res.shape))
pca = joblib.load('h:/static/model/pca.model')
newX = pca.transform(res)
for v in newX:
sr.rpush('final_photos', v.tostring())
logger.info("{}》》》{}成功".format(res.shape, newX.shape))
def getData(dir_path='h:/static/photo', train_len=6, size=(64, 64)):
sr = StrictRedis()
# 清除
sr.delete('deal_photos', 'map')
count = 1
for v in readPhotos(dir_path, train_len):
d_name, file = v
try:
im = Image.open(file, 'r')
# 预处理
data = pre_deal(im, size)
sr.rpush('deal_photos', data.tostring())
name = os.path.basename(file)
sr.rpush('map', os.path.join(d_name, name))
except Exception as e:
logger.warning("file:{} error:{}".format(file, e))
logger.info("成功处理:{:0>5d}-{}-{}-{}".format(count, d_name, data.shape,
data.dtype))
count += 1
def get_photo():
# print(request.headers)
img = request.files.get('file', None)
if img:
im = Image.open(img)
msg = "{} {} {}".format(im.size, im.format, im.mode)
logger.info(msg)
data = np.array(im.resize((32, 32))) / 255.0
datas = np.array([data])
# 定义输入和Label以填充容器，测试时dropout为0
test_feed_dict = {datas_placeholder: datas, dropout_placeholdr: 0}
predicted_index = [(i, v) for i, v in enumerate(
sess.run(logits, feed_dict=test_feed_dict)[0])]
three_index = heapSort(predicted_index,
key=lambda x: x[-1],
count_num=10)
labels = [(sr.lindex('cnn_mapping', int(_[0])).decode('utf-8'), _[1])
for _ in three_index if _[1] > 5]
logger.info("可能的种类:{}".format(labels))
if not labels:
return ''
# 读取图片以及索引
res = list()
mapping = list()
key_score = list()
max_score = 0
for label in labels:
key, score = label
if max_score < score:
max_score = max_score
photos = sr.lrange(key, 0, -1)
res.extend(photos)
key_score.extend(score for _ in range(len(photos)))
mapping.extend(sr.lrange('{}_index'.format(key), 0, -1))
# 计算pca相似度
result = []
for v in use_model(im, pca, res, mapping, key_score, max_score):
paths, kind_score, cos_score = v
path = url_for('static',
filename='baiduPhoto/{}/{}'.format(*paths))
logger.info(paths)
result.append({
'kind_name': paths[0],
'url': path,
'kind_score': kind_score,
'cos_score': cos_score
})
return json.dumps(result, ensure_ascii=False)
return ''
def final_deal():
# 获得图片矩阵
sr = StrictRedis()
res = np.array(
list(
np.fromstring(i, dtype=np.float64)
for i in sr.lrange('deal_photos', 0, -1)))
logger.info('训练矩阵:{}'.format(res.shape))
# # pca降维
pca = PCA(n_components=0.9)
newX = pca.fit_transform(res)
sr.delete('final_photos')
for v in newX:
sr.rpush('final_photos', v.tostring())
logger.info("所占特征百分比:{}".format(np.sum(pca.explained_variance_ratio_)))
logger.info("特征矩阵大小:{}".format(pca.components_.shape))
path = 'H:/static/model'
if not os.path.exists(path):
os.mkdir(path)
# 保存pca模型
joblib.dump(pca, os.path.join(path, 'pca.model'))
fc = tf.layers.dense(flatten, 400, activation=tf.nn.relu)
# 加上DropOut，防止过拟合
dropout_fc = tf.layers.dropout(fc, dropout_placeholdr)
# 未激活的输出层
logits = tf.layers.dense(dropout_fc, 133)
predicted_labels = tf.arg_max(logits, 1)
sess = tf.Session()
# 如果是测试，载入参数
# 用于保存和载入模型
saver = tf.train.Saver()
saver.restore(sess, model_path)
logger.info("从{}载入模型".format(model_path))
# 自动跳转到静态上传网页
@app.route('/')
def hello_world():
return send_file('h:/static/update.html')
@app.route('/getPhoto/', methods=['POST', 'GET'])
def get_photo():
# print(request.headers)
img = request.files.get('file', None)
if img:
im = Image.open(img)
msg = "{} {} {}".format(im.size, im.format, im.mode)