Best way to map Text and Image while loading the data-CodePudding

I have a csv file which looks somewhat like in the photo.

I'm building a model that takes both image and its corresponding text (df['Content'] as input .

I wanted to know the best way to load this data in the following way:

Loading the images from df['Image_location'] into a tensor.
And preserving the order of the image to the corresponding text.
Preserving the corresponding label (df['Sentiment'])

Any ideas on how this can be done?

CodePudding user response：

You can try using the tf.data.Dataset API.

Create dummy data:

import numpy
from PIL import Image

for i in range(1, 3):
  imarray = numpy.random.rand(64,64,3) * 255
  im = Image.fromarray(imarray.astype('uint8')).convert('RGBA')
  im.save('result_image{}.png'.format(i))

Process:

import tensorflow as tf
import pandas as pd
import matplotlib.pyplot as plt

df = pd.DataFrame(data= {'Location': ['some.txt', 'some-other.txt'], 
                         'Content': ['This road was ok', 'This was wonderful'],
                         'Score': [0.0353, -0.341],
                         'Sentiment': ['Neutral', 'Positive'],
                         'Image_location': ['/content/result_image1.png', '/content/result_image2.png']})

features = df[['Content', 'Image_location']]
labels = df['Sentiment']

dataset = tf.data.Dataset.from_tensor_slices((features, labels))
def process_path(x):
  content, image_path = x[0], x[1]
  img = tf.io.read_file(image_path)
  img = tf.io.decode_png(img, channels=3)
  return content, img

dataset = dataset.map(lambda x, y: (process_path(x), y))

for x, y in dataset.take(1):
  content = x[0]
  image = x[1]
  print('Content -->', content)
  print('Sentiment -->', y)
  plt.imshow(image.numpy())

Content --> tf.Tensor(b'This road was ok', shape=(), dtype=string)
Sentiment --> tf.Tensor(b'Neutral', shape=(), dtype=string)