Create set_images.py

set_images.py

@@ -0,0 +1,42 @@
+import json
+from datasets import Dataset, Features, Image, Value
+from pathlib import Path
+
+
+class MyDataset(datasets.GeneratorBasedBuilder):
+    BUILDER_CONFIGS = [{"name": "default", "version": datasets.Version("0.1.0")}]
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            features=Features({
+                "source_language": Value("string"),
+                "target_language": Value("string"),
+                "source_image_file_name": Image(),
+                "target_image_file_name": Image(),
+                "texts": Value("string")
+            })
+        )
+
+    def _split_generators(self, dl_manager):
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.DEV,  
+                gen_kwargs={"filepath": "data/dev/metadata.jsonl", "img_dir": "data/dev"},
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.TEST,
+                gen_kwargs={"filepath": "data/test/metadata.jsonl", "img_dir": "data/test"},
+            ),
+        ]
+
+    def _generate_examples(self, filepath, img_dir):
+        img_dir = Path(img_dir)
+        with open(filepath, encoding="utf-8") as f:
+            for line_num, line in enumerate(f):
+                row = json.loads(line)
+                img_path = img_dir / row["source_image_file_name"]  # Assumes "image": "relative/path.jpg" in JSONL; use full relative if needed (e.g., "images/file.jpg")
+                img_path_target = img_dir / row["target_image_file_name"]
+                if img_path.exists():
+                    yield line_num, {"source_image_file_name": str(img_path), **row}  # Path as str for Image() decoding
+                if img_path_target.exists():
+                    yield line_num, {"target_image_file_name": str(img_path), **row}  # Path as str for Image() decoding