使用 CNN Keras 进行多图像类分类

Question

I am new in the field of CNN.
I want to classify 100 images.
These images are belong to 12 classes.
I have a csv file that contain the image ID (the ID is the image file name) and the image classes.
How I can use and divide the classes from the csv file into train and test.
I mean when I split the data into train and test, how I can divide the 100 pictures and their classes from the csv file to train and test, then feed them to the model.
I use python and Keras library.
Thanks in advance.

import os
import cv2
from keras.layers import Input,Dense,Flatten,Dropout,merge,Reshape,Conv3D,MaxPooling3D,UpSampling3D,Conv2DTranspose
from keras.layers.normalization import BatchNormalization
from keras.models import Model,Sequential
from keras.callbacks import ModelCheckpoint
from keras.optimizers import Adadelta, RMSprop,SGD,Adam
from keras import regularizers
from keras import backend as K
import numpy as np
import scipy.misc
import numpy.random as rng
from PIL import Image, ImageDraw, ImageFont
from sklearn.utils import shuffle
import nibabel as nib #reading MR images
from sklearn.model_selection import train_test_split
import math
import glob
from matplotlib import pyplot as plt
import pandas as pd
from google.colab import drive
drive.mount('/content/drive')
files = glob.glob('/content/drive/My Drive/im_id/*')
files[0]
len(files)
images = []
for f in range(len(files)):
  a = nib.load(files[f])
  a = a.get_data()
  images.append(a)
print(a.shape)
images = np.asarray(images)
print(images.shape)
labeles = pd.read_csv('/content/drive/My 
Drive/img_id.csv')
print(labeles)
class_names = labeles["Class"]
from tensorflow.keras import datasets, layers, models
model = models.Sequential()
model.add(layers.Conv3D(32, (3, 3, 3), activation='relu', input_shape=(110, 110, 110, 1)))
model.add(layers.MaxPooling3D((2, 2, 2)))
model.add(layers.Conv3D(64, (3, 3, 3), activation='relu'))
model.add(layers.MaxPooling3D((2, 2, 2)))
model.add(layers.Conv3D(64, (3, 3, 3), activation='relu'))
model.add(layers.Flatten())
model.add(layers.Dense(64, activation='relu'))
model.add(layers.Dense(10, activation='softmax'))
model.summary()

Answer 1

You need to create a tf.data.Dataset , which is an abstraction that represents a sequence of elements, in which each element consists of one or more components. In other words, it's a generator in python.

I don't know your data, so I can't help you properly. The best I can do is to give you a gist of how to do it and to share some links.

In brief, you need to read your csv file, I suggest using pandas , and the images' path. Now, you can split them using NumPy , for example, you can shuffle your data and then get the first 10% of values in the array for testing and the rest for training.

Next, you need to use the tf.data API to generate the datasets. The method tf.data.Dataset.from_tensor_slices do the job for you. You just need to pass the NumPy arrays as parameters and get the generators to pass to your Keras pipeline. You must do that to your training and testing arrays.

I have a pipeline to do that here . Feel free to use it. You can also check the manual here and here

I hope it helps you.