Upload pm25vision.py with huggingface_hub

pm25vision.py

@@ -1,5 +1,7 @@
+import os
 import csv
 import datasets
+import tarfile
 
 
 class PM25Vision(datasets.GeneratorBasedBuilder):
@@ -17,7 +19,7 @@ class PM25Vision(datasets.GeneratorBasedBuilder):
                     "quality_score": datasets.Value("float64"),
                     "downloaded_at": datasets.Value("string"),
                     "pm25": datasets.Value("float64"),
-                    "filename": datasets.Value("string"),  
+                    "image": datasets.Image(),   # 图片字段
                     "quality": datasets.Value("string"),
                     "pm25_bin": datasets.Value("string"),
                 }
@@ -26,20 +28,56 @@ class PM25Vision(datasets.GeneratorBasedBuilder):
         )
 
     def _split_generators(self, dl_manager):
+        # 下载压缩包
+        train_path = dl_manager.download("train.tar.gz")
+        test_path = dl_manager.download("test.tar.gz")
 
         return [
             datasets.SplitGenerator(
                 name=datasets.Split.TRAIN,
-                gen_kwargs={"filepath": "train/metadata.csv"},
+                gen_kwargs={"archive_path": train_path, "split": "train"},
             ),
             datasets.SplitGenerator(
                 name=datasets.Split.TEST,
-                gen_kwargs={"filepath": "test/metadata.csv"},
+                gen_kwargs={"archive_path": test_path, "split": "test"},
             ),
         ]
 
-    def _generate_examples(self, filepath):
-        with open(filepath, encoding="utf-8") as f:
-            reader = csv.DictReader(f)
-            for i, row in enumerate(reader):
-                yield i, row
+    def _generate_examples(self, archive_path, split):
+        """
+        如果是普通模式：解压 -> 遍历 metadata.csv
+        如果是 streaming 模式：直接读取 tar.gz -> 逐个 yield
+        """
+        if self.config.streaming:
+            # Streaming 模式：用 dl_manager.iter_archive 直接迭代 tar.gz
+            with tarfile.open(archive_path, "r:gz") as tar:
+                members = {m.name: tar.extractfile(m) for m in tar.getmembers() if m.isfile()}
+
+                # 找 metadata.csv
+                meta_file = None
+                for name in members:
+                    if name.endswith("metadata.csv"):
+                        meta_file = members[name]
+                        break
+                if meta_file is None:
+                    raise FileNotFoundError("metadata.csv not found in archive")
+
+                reader = csv.DictReader(line.decode("utf-8") for line in meta_file)
+                for i, row in enumerate(reader):
+                    image_name = os.path.join("images", row["filename"])
+                    if image_name in members:
+                        row["image"] = {"bytes": members[image_name].read(), "path": row["filename"]}
+                        yield i, row
+
+        else:
+            # 默认模式：先解压，再从磁盘读
+            base_dir = datasets.filesystems.extract_archive(archive_path)
+            metadata_file = os.path.join(base_dir, "metadata.csv")
+            images_dir = os.path.join(base_dir, "images")
+
+            with open(metadata_file, encoding="utf-8") as f:
+                reader = csv.DictReader(f)
+                for i, row in enumerate(reader):
+                    image_path = os.path.join(images_dir, row["filename"])
+                    row["image"] = image_path
+                    yield i, row