Update ucf_crime.py

ucf_crime.py

@@ -1,3 +1,4 @@
+import os
 import datasets
 
 
@@ -57,21 +58,21 @@ _DATA_URLS = {
 }
 _SPLIT_PATH = "UCF_Crimes-Train-Test-Split/"
 _SPLIT_FILES = {
-    "action_recognition": {
-        "train": [
-            _SPLIT_PATH + f"Action_Recognition_splits/train_{i:03}.txt"
+    "event_recognition": {
+        "train": {
+            f"event{i}": _SPLIT_PATH + f"Action_Recognition_splits/train_{i:03}.txt"
             for i in range(1, 5)
-        ],
-        "test": [
-            _SPLIT_PATH + f"Action_Recognition_splits/test_{i:03}.txt"
+        },
+        "test": {
+            f"event{i}": _SPLIT_PATH + f"Action_Recognition_splits/test_{i:03}.txt"
             for i in range(1, 5)
-        ],
+        },
     },
     "anomaly_detection": {
         "train": [_SPLIT_PATH + "Anomaly_Detection_splits/Anomaly_Train.txt"],
         "test": [_SPLIT_PATH + "Anomaly_Detection_splits/Anomaly_Test.txt"],
     },
-    "event_recognition": {
+    "temporal_anomaly": {
         "train": [_SPLIT_PATH + "Temporal_Anomaly_Annotation_for_Testing_Videos.txt"],
     },
 }
@@ -100,8 +101,7 @@ class UCFCrime(datasets.GeneratorBasedBuilder):
     BUILDER_CONFIGS = [
         UCFCrimeConfig(name="all"),
         UCFCrimeConfig(name="anomaly"),
-        UCFCrimeConfig(name="action"),
-        UCFCrimeConfig(name="event"),
+        *[UCFCrimeConfig(name=f"event{i}") for i in range(1, 5)],
         UCFCrimeConfig(name="test"),  # For speed-loading
     ]
 
@@ -116,21 +116,66 @@ class UCFCrime(datasets.GeneratorBasedBuilder):
 
     def _split_generators(self, dl_manager):
         """Return SplitGenerators."""
-        data_url = self.config.data_urls
-        if self.config.name == "test":
-            data_url = data_url["test"]
-        data_file = dl_manager.download(data_url)
-        return [
-            datasets.SplitGenerator(
-                name=datasets.Split.TRAIN,
-                # These kwargs will be passed to _generate_examples
-                gen_kwargs={"data_file": dl_manager.iter_archive(data_file)},
-            ),
-        ]
+        data_urls = self.config.data_urls
+        if self.config.name != "test":
+            data_urls.pop("test")
+            if "event" in self.config.name:
+                data_urls.pop("normal-train-part1")
+                data_urls.pop("normal-train-part2")
+                data_urls.pop("normal-test")
+                split_url = _SPLIT_FILES["event_recognition"]
+            elif "anomaly" == self.config.name:
+                data_urls.pop("normal-event-recognition")
+                split_url = _SPLIT_FILES["anomaly_detection"]
+            data_url = data_urls
+        else:
+            data_url = data_urls["test"]
+
+        path = dl_manager.download(data_url)
+        if split_url is not None:
+            split_path = dl_manager.download(split_url)
+        if self.config.name in ("all", "test"):
+            return [
+                datasets.SplitGenerator(
+                    name=datasets.Split.TRAIN,
+                    # These kwargs will be passed to _generate_examples
+                    gen_kwargs={
+                        "extracted_path": dl_manager.extract(path),
+                        "split_path": None,
+                    },
+                ),
+            ]
+        elif self.config.name in ("anomaly", *[f"event{i}" for i in range(1, 5)]):
+            return [
+                datasets.SplitGenerator(
+                    name=datasets.Split.TRAIN,
+                    # These kwargs will be passed to _generate_examples
+                    gen_kwargs={
+                        "extracted_path": dl_manager.extract(path),
+                        "split_path": split_path["train"][self.config.name],
+                    },
+                ),
+                datasets.SplitGenerator(
+                    name=datasets.Split.TEST,
+                    # These kwargs will be passed to _generate_examples
+                    gen_kwargs={
+                        "extracted_path": dl_manager.extract(path),
+                        "split_path": split_path["test"][self.config.name],
+                    },
+                ),
+            ]
 
-    def _generate_examples(self, data_file: str):
+    def _generate_examples(self, extracted_path: str, split_path: dict):
         """Yields example."""
-        for path, file in data_file:
-            if path.endswith(".mp4"):
-                print(file.read())
-        yield None, {"video_path": data_file}
+        idx = 0
+        split_files = []
+        with open(split_path, "r") as f:
+            for line in f.readlines():
+                split_files.append(line.strip().split("/")[-1])
+        for root, _, files in os.walk(extracted_path, topdown=False):
+            relative_dir_path = root.replace(os.path.abspath(extracted_path) + os.sep, "")
+            for name in files:
+                relative_file_path = os.path.join(relative_dir_path, name)
+                if relative_file_path.endswith(".mp4") and name in split_files:
+                    yield idx, {"video_path": relative_file_path}
+                    idx += 1