rock-glacier-dataset.py

"""Rock Glacier dataset with images of the chilean andes"""

import os

import datasets
from datasets.tasks import ImageClassification

_HOMEPAGE = "https://github.com/alcazar90/rock-glacier-detection"


_CITATION = """\
@ONLINE {rock-glacier-dataset,
    author="CMM-Glaciares",
    title="Rock Glacier Dataset",
    month="October",
    year="2022",
    url="https://github.com/alcazar90/rock-glacier-detection"
}
"""

_DESCRIPTION = """\
TODO: Add a description...
"""


_URLS = {
	"train": "https://huggingface.co/datasets/alkzar90/rock-glacier-dataset/resolve/main/data/train.tar.gz",
	"validation": "https://huggingface.co/datasets/alkzar90/rock-glacier-dataset/resolve/main/data/validation.tar.gz"
}

_NAMES = ["glaciar", "cordillera"]


class RockGlacierDataset(datasets.GeneratorBasedBuilder):
   """Rock Glacier images dataset."""

   def _info(self):
       return datasets.DatasetInfo(
       	   description=_DESCRIPTION,
       	   features=datasets.Features(
	       {
                   "image": datasets.Image(),
                   "labels": datasets.features.ClassLabel(names=_NAMES),
	       }
	   ),
	   supervised_keys=("image", "labels"),
	   homepage=_HOMEPAGE,
	   citation=_CITATION,
	   task_templates=[ImageClassification(image_column="image", label_column="labels")],
	)
	

   def _split_generators(self, dl_manager):
       data_files = dl_manager.download_and_extract(_URLS)
       return [
	   datasets.SplitGenerator(
	       name=datasets.Split.TRAIN,
	       gen_kwargs={
		   "files": dl_manager.iter_files([data_files["train"]]),
	       },
	   ),
	   datasets.SplitGenerator(
	       name=datasets.Split.VALIDATION,
	       gen_kwargs={
		   "files": dl_manager.iter_files([data_files["validation"]]),
	       },
 	   ),
       ]

   def _generate_examples(self, files):
       for i, path in enumerate(files):
           file_name = os.path.basename(path)
           if file_name.endswith(".png"):
               yield i, {
                   "image": path,
		   "labels": os.path.basename(os.path.dirname(path)).lower(),
	       }