Make dataset streamable

illustrated_ads.py

@@ -14,10 +14,10 @@
 """Dataset of illustrated and non illustrated 19th Century newspaper ads."""
 
 import ast
+import os
 import pandas as pd
 import datasets
 from PIL import Image
-from pathlib import Path
 
 # TODO: Add BibTeX citation
 # Find for instance the citation on arxiv or on the dataset repo/website
@@ -46,6 +46,19 @@ _LICENSE = "Public Domain"
 
 _URLS = "https://zenodo.org/record/5838410/files/images.zip?download=1"
 
+_DTYPES = {
+    "page_seq_num": "int64",
+    "edition_seq_num": "int64",
+    "batch": "string",
+    "lccn": "string",
+    "score": "float64",
+    "place_of_publication": "string",
+    "name": "string",
+    "publisher": "string",
+    "url": "string",
+    "page_url": "string",
+}
+
 
 # TODO: Name of the dataset usually match the script name with CamelCase instead of snake_case
 class IllustratedAds(datasets.GeneratorBasedBuilder):
@@ -94,48 +107,43 @@ class IllustratedAds(datasets.GeneratorBasedBuilder):
         )
 
     def _split_generators(self, dl_manager):
-        data_dir = dl_manager.download_and_extract(_URLS)
-        return [
-            datasets.SplitGenerator(
-                name=datasets.Split.TRAIN,
-                gen_kwargs={
-                    "data_dir": Path(data_dir),
-                },
-            ),
-        ]
-
-    def _generate_examples(self, data_dir):
-        dtypes = {
-            "page_seq_num": "int64",
-            "edition_seq_num": "int64",
-            "batch": "string",
-            "lccn": "string",
-            "score": "float64",
-            "place_of_publication": "string",
-            "name": "string",
-            "publisher": "string",
-            "url": "string",
-            "page_url": "string",
-        }
+        images = dl_manager.download_and_extract(_URLS)
+        annotations = dl_manager.download(
+            [
+                "https://zenodo.org/record/5838410/files/ads.csv?download=1",
+                "https://zenodo.org/record/5838410/files/sample.csv?download=1"
+            ]
+        )
         df_labels = pd.read_csv(
-            "https://zenodo.org/record/5838410/files/ads.csv?download=1", index_col=0
+            annotations[0], index_col=0
         )
         df_metadata = pd.read_csv(
-            "https://zenodo.org/record/5838410/files/sample.csv?download=1",
+            annotations[1],
             index_col=0,
-            dtype=dtypes,
+            dtype=_DTYPES,
         )
         df_metadata["file"] = df_metadata.filepath.str.replace("/", "_")
         df_metadata = df_metadata.set_index("file", drop=True)
         df = df_labels.join(df_metadata)
         df = df.reset_index()
-        data = df.to_dict(orient="records")
-        for id_, row in enumerate(data):
+        annotations = df.to_dict(orient="records")
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                gen_kwargs={
+                    "images": images,
+                    "annotations": annotations,
+                },
+            ),
+        ]
+
+    def _generate_examples(self, images, annotations):
+        for id_, row in enumerate(annotations):
             box = ast.literal_eval(row["box"])
             row["box"] = box
             row.pop("filepath")
             ocr = " ".join(ast.literal_eval(row["ocr"]))
             row["ocr"] = ocr
             image = row["file"]
-            row["image"] = Image.open(Path(data_dir / image))
+            row["image"] = os.path.join(images, image)
             yield id_, row