Edit dataset loading script

scripts/gen_train_list.py

@@ -19,7 +19,7 @@ if __name__ == "__main__":
 
             try:
                 for video in os.listdir(subdir_path):
-                    f.write(f"{subdir}/{video}\n")
+                    f.write(f"{subdir}/{video.split('.mp4')[0]}\n")
             except FileNotFoundError as e:
                 logger.error(
                     f"Error processing subdir: {subdir_path}: {e.__class__.__name__}"

test.py

@@ -1,4 +0,0 @@
-from datasets import load_dataset
-
-ds = load_dataset("jherng/xd-violence", name="video", split="train")
-print(ds)

xd-violence.py

@@ -1,8 +1,7 @@
-import json
 import urllib.parse
 
 import datasets
-
+import pandas as pd
 
 _CITATION = """\
 @inproceedings{Wu2020not,
@@ -37,7 +36,7 @@ class XDViolenceConfig(datasets.BuilderConfig):
         super(XDViolenceConfig, self).__init__(**kwargs)
 
 
-class Squad(datasets.GeneratorBasedBuilder):
+class XDViolence(datasets.GeneratorBasedBuilder):
     BUILDER_CONFIGS = [
         XDViolenceConfig(
             name="video",
@@ -123,22 +122,56 @@ class Squad(datasets.GeneratorBasedBuilder):
         # Download videos
         # - get URLs
         # - download annotations
-        
-        train_list_fpath = dl_manager.download_and_extract(urllib.parse.urljoin(_URL, "train_list.txt"))
-        test_ann_fpath = dl_manager.download_and_extract(urllib.parse.urljoin(_URL, "test_annotations.txt"))
-        print(f"{train_list_fpath=}")
-        print(f"{test_ann_fpath=}")
+
+        # Download annotations file
+        annotations_paths = {
+            "train": dl_manager.download_and_extract(
+                urllib.parse.urljoin(_URL, "train_list.txt")
+            ),
+            "test": dl_manager.download_and_extract(
+                urllib.parse.urljoin(_URL, "test_annotations.txt")
+            ),
+        }
+
+        # Obtain video URLs
+        video_urls = {
+            "train": pd.read_csv(
+                annotations_paths["train"],
+                header=None,
+                sep=" ",
+                usecols=[0],
+                names=["video_id"],
+            ).apply(lambda x: urllib.parse.urljoin(_URL, f"video/{x}.mp4")),
+            "test": pd.read_csv(
+                annotations_paths["test"],
+                header=None,
+                sep=" ",
+                usecols=[0],
+                names=["video_id"],
+            ).apply(lambda x: urllib.parse.urljoin(_URL, f"test_videos/{x}.mp4")),
+        }
+        print(video_urls["train"])
+        print()
+        print(video_urls["test"])
+
         return [
             datasets.SplitGenerator(
                 name=datasets.Split.TRAIN,
-                gen_kwargs={"filepath": train_list_fpath},
+                gen_kwargs={
+                    "annotation": annotations_paths["train"],
+                    "video": video_urls["train"],
+                },
             ),
             datasets.SplitGenerator(
                 name=datasets.Split.VALIDATION,
-                gen_kwargs={"filepath": test_ann_fpath},
+                gen_kwargs={
+                    "annotation": annotations_paths["test"],
+                    "video": video_urls["test"],
+                },
             ),
         ]
 
-    def _generate_examples(self, filepath):
+    def _generate_examples(self, annotation, video):
         """This function returns the examples in the raw (text) form."""
+
         pass