Update SATIN.py

SATIN.py

@@ -1,189 +1,61 @@
-"""
-
 import datasets
-import os
-import pyarrow.parquet as pq
-from PIL import Image
-from io import BytesIO
-import numpy as np
-import pandas as pd
-
-
-def load_data(data_dir):
-    parquet_file = [file for file in os.listdir(data_dir) if file.endswith('.parquet')][0]
-    print(parquet_file)
-    parquet_path = os.path.join(data_dir, parquet_file)
+from datasets import load_dataset
 
-    parquet_path = data_dir
-    table = pq.read_table(parquet_path)
-
-    for row in table.iterrecords():
-        image_bytes = row['image']
-        image = Image.open(BytesIO(image_bytes))
-        label = row['label']
-        yield image, label
 
+_CONSTITUENT_DATASETS = ['SAT-4', 'SAT-6', 'NASC-TG2', 'WHU-RS19', 'RSSCN7', 'RS_C11', 'SIRI-WHU', 'EuroSAT',
+                         'NWPU-RESISC45', 'PatternNet', 'RSD46-WHU', 'GID', 'CLRS', 'Optimal-31',
+                         'Airbus-Wind-Turbines-Patches', 'USTC_SmokeRS', 'Canadian_Cropland_Dataset',
+                         'Ships-In-Satellite-Imagery', 'Satellite-Images-of-Hurricane-Damage',
+                         'Brazilian_Coffee_Scenes', 'Brazilian_Cerrado-Savanna_Scenes', 'Million-AID',
+                         'UC_Merced_LandUse_MultiLabel', 'MLRSNet_MultiLabel', 'AID_MultiLabel',
+                         'MultiScene', 'RSI-CB256']
 
 
 class SATINConfig(datasets.BuilderConfig):
+    """BuilderConfig for SATIN"""
 
-
-    def __init__(self, name, description, data_url, class_names, **kwargs):
-
-        Args:
-          data_url: `string`, url to download the zip file from.
-          metadata_urls: dictionary with keys 'train' and 'validation' containing the archive metadata URLs
-          **kwargs: keyword arguments forwarded to super.
+    def __init__(self, name, **kwargs):
 
         super(SATINConfig, self).__init__(version=datasets.Version("1.0.0"), **kwargs)
         self.name = name
-        self.data_url = data_url
-        self.description = description
-        self.class_names = class_names
-
-
-class SATIN(datasets.GeneratorBasedBuilder):
-    SATIN Images dataset
-
-    _SAT_4_NAMES = ['barren land', 'grassland', 'other', 'trees']
-    _SAT_6_NAMES = ['barren land', 'building', 'grassland', 'road', 'trees', 'water']
-
-    BUILDER_CONFIGS = [
-        SATINConfig(
-            name="SAT_4",
-            description="SAT_4.",
-            data_url="https://huggingface.co/datasets/jonathan-roberts1/SAT-4/tree/main/data/",#train-00000-of-00001-e2dcb38bc165dfb0.parquet",
-            class_names = _SAT_4_NAMES
-            #metadata_urls={
-            #    "train": "https://link-to-breakfast-foods-train.txt",
-         ),
-        SATINConfig(
-            name="SAT_6",
-            description="SAT_6.",
-            data_url="https://huggingface.co/datasets/jonathan-roberts1/SAT-6/tree/main/data/",#train-00000-of-00001-c47ada2c92f814d2.parquet",
-            class_names = _SAT_6_NAMES
-        )
-    ]
-
-    @property
-    def url_prefix(self):
-        return {
-            "SAT-4": "https://huggingface.co/datasets/jonathan-roberts1/SAT-4/tree/main/data/",#train-00000-of-00001-e2dcb38bc165dfb0.parquet",#train-00000-of-00001-e2dcb38bc165dfb0.parquet",
-            "SAT-6": "https://huggingface.co/datasets/jonathan-roberts1/SAT-6/tree/main/data/",
-        }
-
-    def _info(self):
-        return datasets.DatasetInfo(
-        description=self.config.description,
-        features=datasets.Features(
-            {
-                "image": datasets.Image(),
-                "label": datasets.ClassLabel(names=self.config.class_names),
-            }
-        ),
-        supervised_keys=("image", "label"),
-        #homepage=_HOMEPAGE,
-        #citation=_CITATION,
-        #license=_LICENSE,
-        #task_templates=[ImageClassification(image_column="image", label_column="label")],
-    )
-
-    def _split_generators(self, dl_manager):
-        url = self.config.data_url
-        data_dir = dl_manager.download_and_extract(url)#, use_auth_token=True)
-        print(data_dir)
-        return [
-            datasets.SplitGenerator(
-                name=datasets.Split.TRAIN,
-                gen_kwargs={"data_dir": data_dir},
-            ),
-        ]
-
-    def _generate_examples(self, data_dir):
-        #base_url = self.url_prefix[self.config.name]
-        file_url = self.config.data_url
-        use_auth_token = os.environ.get("HUGGINGFACE_TOKEN")
-
-        with NamedTemporaryFile() as file:
-            download(file_url, file.name, use_auth_token=use_auth_token)
-            df = pd.read_parquet(file.name)
-
-        for idx, row in df.iterrows():
-            example = {
-                "image": row["image"],
-                "label": row["label"],
-            }
-            yield idx, example
-
-
-    #def _generate_examples(self, data_dir):
-    #    for idx, (image, label) in enumerate(load_data(data_dir)):
-    #        image_array = np.array(image)
-    #        yield idx, {"image": image_array, "label": label}
-"""
-
-
-from datasets.utils.download_manager import DownloadManager
-import tempfile
-import datasets
-import os
-import pyarrow.parquet as pq
-from PIL import Image
-from io import BytesIO
-import numpy as np
-import pandas as pd
-
-
-class SATINConfig(datasets.BuilderConfig):
-
-
-    def __init__(self, name, description, data_url, class_names, **kwargs):
-
-        super(SATINConfig, self).__init__(version=datasets.Version("1.0.0"), **kwargs)
-        self.name = name
-        self.data_url = data_url
-        self.description = description
-        self.class_names = class_names
+        self.hf_dataset_name = 'jonathan-roberts1' + "/" + name
+        self.description = None
+        self.features = None
 
+        #stream_dataset_info = load_dataset(self.hf_dataset_name, streaming=True, split='train').info
+        #self.description = stream_dataset_info.description
+        #self.features = stream_dataset_info.features
 
 
 class SATIN(datasets.GeneratorBasedBuilder):
     """SATIN Images dataset"""
 
-    _SAT_4_NAMES = ['barren land', 'grassland', 'other', 'trees']
-    _SAT_6_NAMES = ['barren land', 'building', 'grassland', 'road', 'trees', 'water']
-
-    BUILDER_CONFIGS = [
-        SATINConfig(
-            name="SAT_4",
-            description="SAT_4.",
-            data_url="jonathan-roberts1/SAT-4",#https://huggingface.co/datasets/jonathan-roberts1/SAT-4/blob/main/data/train-00000-of-00001-e2dcb38bc165dfb0.parquet?raw=true",
-            class_names=_SAT_4_NAMES
-        ),
-        SATINConfig(
-            name="SAT_6",
-            description="SAT_6.",
-            data_url="jonathan-roberts1/SAT-6",#"https://huggingface.co/datasets/jonathan-roberts1/SAT-6/blob/main/data/train-00000-of-00001-c47ada2c92f814d2.parquet?raw=true",
-            class_names=_SAT_6_NAMES
-        )
-    ]
+    BUILDER_CONFIGS = [SATINConfig(name=dataset_name) for dataset_name in _CONSTITUENT_DATASETS]
+
+    """
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.config.hf_dataset = load_dataset(self.config.hf_dataset_name)
+        self.config.description = self.config.hf_dataset['train'].description
+        self.config.features = self.config.hf_dataset['train'].features
+        print(self.config.features)
+    """
 
     def _info(self):
+        if self.config.description is None or self.config.features is None:
+            stream_dataset_info = load_dataset(self.config.hf_dataset_name, streaming=True, split='train').info
+            self.config.description = stream_dataset_info.description
+            self.config.features = stream_dataset_info.features
+        print(f'info {self.config.features}')
         return datasets.DatasetInfo(
             description=self.config.description,
-            features=datasets.Features(
-                {
-                    "image": datasets.Image(),
-                    "label": datasets.ClassLabel(names=self.config.class_names),
-                }
-            ),
-            supervised_keys=("image", "label"),
+            features=self.config.features,
+            #supervised_keys=("image", "label"),
         )
 
+
     def _split_generators(self, dl_manager):
-        #data_path = dl_manager.download(self.config.data_url)
-        from datasets import load_dataset
-        dataset = load_dataset(self.config.data_url)
+        dataset = load_dataset(self.config.hf_dataset_name)
         return [
             datasets.SplitGenerator(
                 name=datasets.Split.TRAIN,
@@ -193,9 +65,11 @@ class SATIN(datasets.GeneratorBasedBuilder):
 
     def _generate_examples(self, data_path):
         # iterate over the Huggingface dataset and yield the idx, image and label
-        huggingface_dataset = data_path["train"]
+        _DEFAULT_SPLIT = 'train'
+        huggingface_dataset = data_path['train']
+        features = huggingface_dataset.features
         for idx, row in enumerate(huggingface_dataset):
-            yield idx, {"image": row["image"], "label": row["label"]}
+            yield idx, {feature: row[feature] for feature in features}