Folder structure for TensorFlow image segmentation?

I am using this tutorial to do image segmentation with tensorflow. However, I want to make my own data for this. What should my folder structure be if I'm trying to stay consistent with the tutorial? Currently, I have the following:

path/to/image
  train/
    image/
      1.png
      ...
    segmentation_mask/
      1.jpg
  test/
    image/
      6.png
      ...
    segmentation_mask/
      6.jpg
      ...

I'm trying to read the dataset using tf.data_folder.ImageFolder("path/to/image") and then .as_dataset() to convert it to be a dataset. However, this raises errors throughout the rest of the program. So, how to I structure my folders and how do I load the dataset in properly so there would not be any more errors?

To load your dataset properly and avoid errors, you need to structure your folders as follows:

path/to/image/
  train/
    images/
      1.png
      2.png
      3.png
      ...
    masks/
      1.png
      2.png
      3.png
  test/
    images/
      6.png
      7.png
      8.png
    masks/
      6.png
      7.png
      8.png

Make sure that your image files and corresponding segmentation mask files have the same names and are placed in their respective folders. For example, 1.png in the train/images folder should have its corresponding mask 1.png in the train/masks folder.

To load the dataset using tf.data.Dataset.from_folder(), you can do the following:

import tensorflow as tf

# Define the paths to the train and test directories
train_path = "path/to/image/train"
test_path = "path/to/image/test"

# Define the image and mask subdirectories
image_dir = "images"
mask_dir = "masks"

# Load the train dataset
train_dataset = tf.data.Dataset.from_folder(
    train_path,
    image_size=(height, width),  # Set the desired image size
    seed=123,  # Set a seed for reproducibility (optional)
    labels="inferred",
    label_mode=None,  # No labels as we are loading masks
    image_folder=image_dir,
    mask_folder=mask_dir,
    interpolation="nearest"  # Set the interpolation method for resizing (optional)
)

# Load the test dataset
test_dataset = tf.data.Dataset.from_folder(
    test_path,
    image_size=(height, width),  # Set the desired image size
    seed=123,  # Set a seed for reproducibility (optional)
    labels="inferred",
    label_mode=None,  # No labels as we are loading masks
    image_folder=image_dir,
    mask_folder=mask_dir,
    interpolation="nearest"  # Set the interpolation method for resizing (optional)
)

Let me know if this works for you!