xception-foodlg

agent.py

# Licensed to the Apache Software Foundation (ASF) under one
# or more contributor license agreements.  See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership.  The ASF licenses this file
# to you under the Apache License, Version 2.0 (the
# "License"); you may not use this file except in compliance
# with the License.  You may obtain a copy of the License at
#
#   http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
# KIND, either express or implied.  See the License for the
# specific language governing permissions and limitations
# under the License.
# =============================================================================
from builtins import str
from builtins import object

from multiprocessing import Process, Queue
from flask import Flask,request, send_from_directory, jsonify
from flask_cors import CORS, cross_origin
import os, traceback, sys
import time
from werkzeug.utils import secure_filename
from werkzeug.datastructures import CombinedMultiDict, MultiDict
import pickle
import uuid

class MsgType(object):
   def __init__(self, name):
       self.name = name
   def __str__(self):
       return self.name
   def __repr__(self):
       return "<Msg: %s>" % self
   def equal(self,target):
       return str(self) == str(target)

   def is_info(self):
       return self.name.startswith('kInfo')
   def is_command(self):
       return self.name.startswith('kCommand')
   def is_status(self):
       return self.name.startswith('kStatus')
   def is_request(self):
       return self.name.startswith('kRequest')
   def is_response(self):
       return self.name.startswith('kResponse')

   @staticmethod
   def parse(name):
       return getattr(MsgType,str(name))
   @staticmethod
   def get_command(name):
       if name=='stop':
           return MsgType.kCommandStop
       if name=='pause':
           return MsgType.kCommandPause
       if name=='resume':
           return MsgType.kCommandResume
       return MsgType.kCommand

types =  ['kInfo','kInfoMetric',
           'kCommand','kCommandStop','kCommandPause','kCommandResume',
           'kStatus','kStatusRunning','kStatusPaused','kStatusError',
           'kRequest','kResponse']

for t in types:
    setattr(MsgType,t,MsgType(t))

#####   NOTE the server currently only can handle request sequentially

app = Flask(__name__)
top_k_=5

class Agent(object):

    def __init__(self,port):
        info_queue = Queue()
        command_queue = Queue()
        self.p = Process(target=start, args=(port, info_queue,command_queue))
        self.p.start()
        self.info_queue=info_queue
        self.command_queue=command_queue
        return

    def pull(self):
        if not self.command_queue.empty():
            msg,data=self.command_queue.get()
            if msg.is_request():
                data = pickle.loads(data)
            return msg,data
        return None,None

    def push(self,msg,value):
        self.info_queue.put((msg,value))
        return

    def stop(self):
        #sleep a while, wait for http response finished
        time.sleep(1)
        self.p.terminate()

def start(port,info_queue,command_queue):
    global info_queue_, command_queue_, data_
    info_queue_=info_queue
    command_queue_=command_queue
    data_ = []
    app.run(host='0.0.0.0', port=port)
    return

def getDataFromInfoQueue(need_return=False):
    global info_queue_, data_
    if not need_return:
        while not info_queue_.empty():
            msg,d = info_queue_.get()
            data_.append(d)
    else:
        while True: # loop until get answer
            while not info_queue_.empty():
                msg,d = info_queue_.get()
                if msg.is_info():
                    data_.append(d)
                else:
                    return msg,d
            time.sleep(0.01)

@app.route("/")
@cross_origin()
def index():
    try:
        req=send_from_directory(os.getcwd(),"index.html", mimetype='text/html')
    except:
        traceback.print_exc()
        return "error"
    return req

# support two operations for user to monitor the training status
@app.route('/getAllData')
@cross_origin()
def getAllData():
    global data_
    try:
        getDataFromInfoQueue()
    except:
        traceback.print_exc()
        return failure("Internal Error")
    return success(data_)


@app.route('/getTopKData')
@cross_origin()
def getTopKData():
    global data_
    try:
        k = int(request.args.get("k", top_k_))
    except:
        traceback.print_exc()
        return failure("k should be integer")
    try:
        getDataFromInfoQueue()
    except:
        traceback.print_exc()
        return failure("Internal Error")
    return success(data_[-k:])

@app.route("/api", methods=['POST'])
@cross_origin()
def api():
    global info_queue_,command_queue_
    try:
        files=transformFile(request.files)
        values = CombinedMultiDict([request.args,request.form,files])
        req_str = pickle.dumps(values)
        command_queue_.put((MsgType.kRequest,req_str))
        msg,response=getDataFromInfoQueue(True)
        deleteFiles(files)
        return response
    except:
        traceback.print_exc()
        return failure("Internal Error")

@app.route("/command/<name>", methods=['GET','POST'])
@cross_origin()
def command(name):
    global info_queue_,command_queue_
    try:
        command=MsgType.get_command(name)
        command_queue_.put((command,""))
        msg,response=getDataFromInfoQueue(True)
        return response
    except:
        traceback.print_exc()
        return failure("Internal Error")

def success(data=""):
    '''return success status in json format'''
    res = dict(result="success", data=data)
    return jsonify(res)
def failure(message):
    '''return failure status in json format'''
    res = dict(result="message", message=message)
    return jsonify(res)


def transformFile(files):
    result= MultiDict([])
    for f in files:
        file = files[f]
        unique_filename = str(uuid.uuid4())+secure_filename(file.filename)
        filepath=os.path.join(os.getcwd(), unique_filename)
        file.save(filepath)
        result.add(f, filepath)
    return result


def deleteFiles(files):
    for f in files:
        filepath = files[f]
        os.remove(filepath)
    return

serve.py

#

train.py

## Xception model with download automatically

from keras.applications.xception import Xception
from keras.layers import GlobalAveragePooling2D
from keras.preprocessing.image import ImageDataGenerator
from keras.models import Sequential
from keras.optimizers import SGD
from keras.layers import Dense
from keras.callbacks import ModelCheckpoint, EarlyStopping
from keras.models import Model

import sys
import argparse
import traceback
from agent import MsgType, Agent


def main(args, agent):
    img_rows, img_cols = 299, 299 # Resolution of inputs
    channel = 3
    num_classes = 100
    batch_size = 8
    nb_epoch = 100
#model = inception_v3_model(img_rows, img_cols, channel, num_classes)
    base_model = Xception(weights='imagenet', include_top=False)
# add a global spatial average pooling layer
    x = base_model.output
    x = GlobalAveragePooling2D()(x)
# let's add a fully-connected layer
    x = Dense(1024, activation='relu')(x)
# and a logistic layer -- let's say we have 200 classes
    predictions = Dense(num_classes, activation='softmax')(x)
    model = Model(inputs=base_model.input, outputs=predictions)

    model.compile(optimizer=SGD(lr=1e-3, decay=1e-6, nesterov=True, momentum=0.9), loss='categorical_crossentropy', metrics=['accuracy'])


# (train_images, train_labels) = load_train_data((img_rows, img_cols))
    filepath = "./models/xception-{epoch:02d}-{loss:0.3f}-{acc:0.3f}-{val_loss:0.3f}-{val_acc:0.3f}.hdf5"
    checkpoint = ModelCheckpoint(filepath, monitor = "loss", verbose = 1, save_best_only = True, mode = 'min')
    early_stopping = EarlyStopping(monitor = "val_loss", patience = 3)
    callbacks_list = [checkpoint, early_stopping]

    datagen = ImageDataGenerator(
        horizontal_flip=True,
        fill_mode="nearest",
        zoom_range=0.3,
        width_shift_range=0.3,
        height_shift_range=0.3,
        rotation_range=30)
    train_gen = train_datagen.flow_from_directory(args.train_data, batch_size=batch_size)
    val_gen = train_datagen.flow_from_directory(args.val_data, batch_size=batch_size)
    for epoch in range(nb_epoch):
        hist = model.fit_generator(gen,
            steps_per_epoch=len(gen) / batch_size,
            epochs=1, verbose=1, validation_data=val_gen,
            validation_steps=len(val_gen) / batch_size,
            callbacks=callbacks_list
            )
        agent.push(MsgType.kInfoMetric, hist[-1])

if __name__ == '__main__':
    try:
        parser = argparse.ArgumentParser()
        action = parser.add_mutually_exclusive_group(required=True)
        action.add_argument('--train', help='Train a model', action='store_true')
        action.add_argument('--test', help='Predict using a saved model', metavar='MODEL')
        action.add_argument('--extract', help='Extract features using a saved model', metavar='MODEL')
        action.add_argument('--port', type=int, default=8333)
        args = parser.parse_args()
        port = args.port
        agent = Agent(port)
        main(args, agent)
        agent.stop()
    except:
        traceback.print_exc()
        sys.stderr.write("  for help use --help \n\n")