[DOCKER] Identify the name from the flower image with keras (tensorflow)
Convolutional Neural Network (CNN) A convolutional neural network Let's experiment to identify the name from the image of the flower!
Introduction
This time, we will put the program on the docker environment.
-
- Environmental preparation
- program
-
- Experiment Place the program in Git.
Environmental preparation
Development environment
Macbook pro 13 2016
install docker
You can easily install docker as a binary from the following site.
https://docs.docker.com/docker-for-mac/
Build a machine learning environment from a Miniconda image
You can install compiled binaries packed with machine learning packages Build an environment with conda (Miniconda).
① Download template
$ docker pull continuumio/miniconda3
② Create Dockerfile
$ vi Dockerfile
:
$ cat Dockerfile
FROM continuumio/miniconda3
RUN apt-get update
RUN pip install --upgrade pip
RUN pip install tensorflow \
pandas \
scipy \
Pillow \
keras
RUN pip install -U scikit-learn
RUN mkdir /home/src/
③ docker build (image creation)
$ docker build -t conda:init .
④ Confirmation
$ docker images
REPOSITORY TAG IMAGE ID CREATED SIZE
conda init daf923e29da8 2 hours ago 1.55 GB
⑤ Start docker
This time, mount the folder to be developed on docker.
# [mount dir]Please rewrite and use.
$ docker run -i -t -v [mount dir]:/home/src conda:init /bin/bash
You can exit by pressing "Ctrl + p" or "Ctrl + q".
When accessing the container again
$ docker ps
CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
745860b72c67 conda:init "/usr/bin/tini -- ..." 11 hours ago Up 11 hours friendly_galileo
#Specify CONTAINER ID
$ docker attach 745860b72c67
This completes the environment preparation
Flower image classification program from here
① Preparation of training data
Collect flower images from the internet! Give the name of the flower to the directory name as shown in the capture below, and place the image of the flower in it.
- I think that there are rights to images in this area, so please collect them yourself.
② Creation of learning model
Prepare a program to save an array of image data and categories.
data_create.py
from PIL import Image
import sys
from sklearn import cross_validation
import os, glob
import numpy as np
import random, math
#
#Generate training data
#
class DataCreate :
def __init__(self, script_name):
Image.LOAD_TRUNCATED_IMAGES = True
def create(self) :
input_dir = "images"
categorys = []
dir_list = os.listdir(input_dir)
for index, dir_name in enumerate(dir_list):
if dir_name == '.DS_Store' :
continue
categorys.append(dir_name)
image_size = 50
train_data = [] #image data,Label data
for idx, category in enumerate(categorys):
try :
print("---", category)
image_dir = input_dir + "/" + category
files = glob.glob(image_dir + "/*.jpg ")
for i, f in enumerate(files):
img = Image.open(f)
img = img.convert("RGB")
img = img.resize((image_size, image_size))
data = np.asarray(img)
train_data.append([data, idx])
except:
print("SKIP : " + category)
#Shuffle data
random.shuffle(train_data)
X, Y = [],[]
for data in train_data:
X.append(data[0])
Y.append(data[1])
test_idx = math.floor(len(X) * 0.8)
xy = (np.array(X[0:test_idx]), np.array(X[test_idx:]),
np.array(Y[0:test_idx]), np.array(Y[test_idx:]))
np.save("./npy/flower", xy)
if __name__ == "__main__":
args = sys.argv
datacreate = DataCreate(args[0])
datacreate.create()
Next, prepare a program to create a learning model.
train.py
import sys
import os
import numpy as np
import pandas as pd
import gc
from keras.models import Sequential
from keras.layers import Convolution2D, MaxPooling2D
from keras.layers import Activation, Dropout, Flatten, Dense
from keras.utils import np_utils
#
#Generate model
#
class TrainModel :
def __init__(self):
input_dir = 'images'
self.nb_classes = len([name for name in os.listdir(input_dir) if name != ".DS_Store"])
x_train, x_test, y_train, y_test = np.load("./npy/flower.npy")
#Normalize the data
self.x_train = x_train.astype("float") / 256
self.x_test = x_test.astype("float") / 256
self.y_train = np_utils.to_categorical(y_train, self.nb_classes)
self.y_test = np_utils.to_categorical(y_test, self.nb_classes)
def train(self, input=None) :
model = Sequential()
# K=32, M=3, H=3
if input == None :
model.add(Convolution2D(32, 3, 3, border_mode='same', input_shape=self.x_train.shape[1:]))
else :
model.add(Convolution2D(32, 3, 3, border_mode='same', input_shape=input))
# K=64, M=3, H=3 (adjustment)
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Convolution2D(64, 3, 3, border_mode='same'))
model.add(Activation('relu'))
# K=64, M=3, H=3 (adjustment)
model.add(Convolution2D(64, 3, 3))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(512))
# biases nb_classes
model.add(Activation('relu'))
model.add(Dropout(0.5))
model.add(Dense(self.nb_classes))
model.add(Activation('softmax'))
model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])
if input == None :
#Learn and save the model
model.fit(self.x_train, self.y_train, batch_size=32, nb_epoch=10)
hdf5_file = "./model/flower-model.hdf5"
model.save_weights(hdf5_file)
#model test
score = model.evaluate(self.x_test, self.y_test)
print('loss=', score[0])
print('accuracy=', score[1])
return model
if __name__ == "__main__":
args = sys.argv
train = TrainModel()
train.train()
gc.collect()
Learning is now complete.
Then try running it.
- It will take some time depending on the specifications.
#Arrange images
$ python data_create.py
---Ai (Tadeai)
---Aogiri
---Aomoria thistle (Okino thistle)
---Agapanthus (Murasaki Clivia)
---Akaza
---Akashouma
---Akabana buckwheat
---Akabana Mitsumata
---Rehmannia glutinosa
---Acanthus(Bear's breeches)
---Aginashi
---Akizaki snowflake
---Akino Unagitsukami
---Salvia japonica (Salvia japonica)
---Akinonogeshi
---Akinowasu Regusa
：
：
#Learning
$ python train.py
:
loss= 0.0410282239616
accuracy= 0.992773946126
The hit rate of 99% is buggy (laughs)
③ Model test for the time being
Prepare a simple program for model testing.
check.py
import train as train
import sys, os
from PIL import Image
import numpy as np
import pandas as pd
if len(sys.argv) <= 1:
quit()
image_size = 50
input_dir = 'images'
categories = [name for name in os.listdir(input_dir) if name != ".DS_Store"]
X = []
for file_name in sys.argv[1:]:
img = Image.open(file_name)
img = img.convert("RGB")
img = img.resize((image_size, image_size))
in_data = np.asarray(img)
X.append(in_data)
X = np.array(X)
model = train.TrainModel().train(X.shape[1:])
model.load_weights("./model/flower-model.hdf5")
predict = model.predict(X)
for pre in predict:
y = pre.argmax()
print("Flower name: ", categories[y])
Test run
Tested with cucumber flowers.
$ python check.py test/flower.jpg
Using TensorFlow backend.
Flower name:cucumber
Cucumber is successful! But there were times when other photos didn't work. I will study a little more and give it again.
Finally
I will also give you how to make a chatbot with learning when you have time.
Recommended Posts