How to Create Dataset for Tensorflow with Images and Labels

Question

I’d like to create a dataset from images in a special folder. Next I’d like to classify them using tensorflow according to the scheme from the dataset.

Is there any quick and efficient way to create a dataset from images and labels?

Answer 1

Your question is a bit vague but here is what I think you mean and you can solve your problem.

I guess you have some images, who belong to certain classes. The images of the same class are in the same folder. Like so:

images/
   cats/
       img_0.png
       img_1.png
       img_2.png
       ...
   dogs/
       img_0.png
       img_1.png
       img_2.png
       ...

You now want a dataset where the x-values are the images and the y-values are the classes.

As described here you can use

train_ds = tf.keras.preprocessing.image_dataset_from_directory(
  data_dir,
  validation_split=0.2,
  subset="training",
  seed=123,
  image_size=(img_height, img_width),
  batch_size=batch_size)

to load your data. In the example I've given the variable for data_dir should hold the value "images". This is a fast and convenient way to load data in TensorFlow. I recommend to click on the link I provided to see the full tutorial on loading data into TensorFlow.