[Script] Download script

EgoSpeak.py

@@ -0,0 +1,59 @@
+import os
+
+from datasets import DatasetInfo, Features, GeneratorBasedBuilder, Split, Value
+
+
+class EgoSpeak(GeneratorBasedBuilder):
+    """EgoSpeak Dataset with EasyCom, Ego4D, and YTConv subdirectories."""
+
+    VERSION = "1.0.0"
+
+    def _info(self):
+        return DatasetInfo(
+            description="EgoSpeak dataset containing precomputed features from EasyCom, Ego4D, and YTConv.",
+            features=Features(
+                {
+                    "sub_dir": Value("string"),
+                    "src_video_filename": Value("string"),
+                    "file_path": Value("string"),
+                }
+            ),
+            homepage="https://huggingface.co/datasets/kjunh/EgoSpeak",
+            license="CC BY-NC 4.0",
+            citation="""\
+@article{egospeak,
+  title={...}
+  author={NAME},
+  journal={ArXiv/Conference},
+  year={2024}
+}
+""",
+        )
+
+    def _split_generators(self, dl_manager):
+        # Path to the dataset directory (change if different)
+        data_dir = os.path.abspath(dl_manager.manual_dir)
+        return [
+            SplitGenerator(
+                name=Split.TRAIN,
+                gen_kwargs={"data_dir": data_dir},
+            ),
+        ]
+
+    def _generate_examples(self, data_dir):
+        """Generate examples from the dataset directory structure."""
+        idx = 0
+        for root, dirs, files in os.walk(data_dir):
+            # Filter for .npy files
+            for file_name in files:
+                if file_name.endswith(".npy"):
+                    file_path = os.path.join(root, file_name)
+                    sub_dir = os.path.relpath(root, data_dir)
+                    src_video_filename = file_name.replace(".npy", "")
+
+                    yield idx, {
+                        "sub_dir": sub_dir,
+                        "src_video_filename": src_video_filename,
+                        "file_path": file_path,
+                    }
+                    idx += 1