Create kanjivg_klee.py

kanjivg_klee.py

@@ -0,0 +1,55 @@
+import datasets
+
+_CITATION = """\
+@InProceedings{huggingface:dataset,
+title = {KanjiVG & Klee},
+author={Kevin Li},
+year={2022}
+}
+"""
+_DESCRIPTION = """\
+Horizontally concatenated 256x128 PNG images of kanjis in two fonts.
+The left kanji is in the KanjiVG font with stroke order indicated by colors.
+See https://github.com/KanjiVG/kanjivg for more info.
+The right kanji is in the Klee font designed for Japanese
+"""
+_HOMEPAGE = "https://huggingface.co/datasets/AlienKevin/kanjivg_klee"
+_LICENSE = "CC0"
+_REPO = "https://huggingface.co/datasets/AlienKevin/kanjivg_klee"
+
+class ImageSet(datasets.GeneratorBasedBuilder):
+    def _info(self):
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=datasets.Features(
+                {
+                    'character': datasets.Value("string"),
+                    'image': datasets.Image(),
+                }
+            ),
+            supervised_keys=None,
+            homepage=_HOMEPAGE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        images_archive = dl_manager.download(f"{_REPO}/resolve/main/images.tgz")
+        image_iters = dl_manager.iter_archive(images_archive)
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                gen_kwargs={
+                    "images": image_iters
+                }
+            ),
+        ]
+
+    def _generate_examples(self, images):
+        idx = 0
+        for filepath, image in images:
+            character = filepath.split('/')[-1][:-4]
+            yield idx, {
+                "image": {"path": filepath, "bytes": image.read()},
+                "character": character,
+            }
+            idx += 1