Create dataset.py

dataset.py

@@ -0,0 +1,51 @@
+import json
+import os
+import tarfile
+
+from datasets import DatasetInfo, GeneratorBasedBuilder, Split, SplitGenerator, Value, Features
+
+class IrrisightDataset(GeneratorBasedBuilder):
+    BUILDER_CONFIGS = [
+        DatasetInfo(
+            description="IRRISIGHT dataset with compressed tar.gz archives",
+            features=Features({
+                "image_path": Value("string"),
+                "label_type": Value("string"),
+                "text_prompt": Value("string"),
+                "polygon": Value("string"),
+                "crs": Value("string"),
+                "split": Value("string")
+            }),
+        )
+    ]
+
+    def _split_generators(self, dl_manager):
+        root_dir = dl_manager.download_and_extract({
+            "maryland": {
+                "data": "https://huggingface.co/datasets/Nibir/IRRISIGHT/resolve/main/New Jersey/2023.tar.gz",
+                "meta": "https://huggingface.co/datasets/Nibir/IRRISIGHT/resolve/main/New Jersey/metadata.jsonl"
+            }
+        })
+
+        return [
+            SplitGenerator(
+                name=Split.TRAIN,
+                gen_kwargs={
+                    "metadata_path": root_dir["maryland"]["meta"],
+                    "image_dir": os.path.join(root_dir["maryland"]["data"], "2023")
+                }
+            )
+        ]
+
+    def _generate_examples(self, metadata_path, image_dir):
+        with open(metadata_path, "r") as f:
+            for idx, line in enumerate(f):
+                record = json.loads(line)
+                yield idx, {
+                    "image_path": os.path.join(image_dir, os.path.basename(record["image_path"])),
+                    "label_type": record.get("label_type", ""),
+                    "text_prompt": record.get("text_prompt", ""),
+                    "polygon": record.get("polygon", ""),
+                    "crs": record.get("crs", ""),
+                    "split": record.get("split", "")
+                }