tf.keras.utils.image_dataset_from_directory can't find images

Question

I have simple code:

data_dir_str = "./photo/pults/samsung_small"
data_dir = pathlib.Path(data_dir_str)
image_count = len(list(data_dir.glob('*.*')))
print('images in directory: ' + str(image_count)) # 12 files in directory

sams_pults = list(data_dir.glob('*.jpg'))
img = PIL.Image.open(str(sams_pults[0]))
img_width, img_height = img.size
#img.show()  # This works well - shows image

batch_size = 10

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

All images are in jpg-format. I can see a picture when use img.show . But image_dataset_from_directory gives me the error: "No images found in directory {directory}. Allowed formats: {ALLOWLIST_FORMATS}"

I don't understand what's wrong.

Answer 1

I have found the solution:

./photo/pults/samsung_small -> ./photo/pults

The path needs to be set not as a dir where images are placed but one level above. Because the dir name is used for classification.