upd: new try

rwth_phoenix_weather_2014.py

@@ -83,6 +83,9 @@ class RWTHPhoenixWeather2014(datasets.GeneratorBasedBuilder):
             "transcription": datasets.Value("string"),
         }
 
+        if self.config.name != "pre-training":
+            features_dict["frames"] = datasets.Sequence(feature=datasets.Image())
+
         return datasets.DatasetInfo(
             description=_DESCRIPTION + self.config.description,
             features=datasets.Features(features_dict),
@@ -96,6 +99,7 @@ class RWTHPhoenixWeather2014(datasets.GeneratorBasedBuilder):
     def _split_generators(self, dl_manager: datasets.DownloadManager):
         example_ids = {}
         annotations = {}
+        frames = {}
 
         dataDirMapper = {
             datasets.Split.TRAIN: "train",
@@ -118,12 +122,23 @@ class RWTHPhoenixWeather2014(datasets.GeneratorBasedBuilder):
             example_ids[split] = df['id']
             annotations[split] = df['annotation']
 
+            frame_archive_urls = dl_manager.download([
+                f"{base_url}/features/fullFrame-210x260px/{dataDirMapper[split]}/{id}.tar"
+                for id in example_ids[split]
+            ])
+
+            frames[split] = [
+                dl_manager.iter_archive(url)
+                for url in frame_archive_urls
+            ]
+
         return [
             datasets.SplitGenerator(
                 name=split,
                 gen_kwargs={
                     "example_ids": example_ids[split],
                     "annotations": annotations[split],
+                    "frames": frames[split],
                 },
             )
             for split in [
@@ -133,11 +148,16 @@ class RWTHPhoenixWeather2014(datasets.GeneratorBasedBuilder):
             ]
         ]
 
-    def _generate_examples(self, example_ids, annotations):
-        for key, (idx, annotation) in enumerate(zip(example_ids, annotations)):
+    def _generate_examples(self, example_ids, annotations, frames):
+        for key, (idx, annotation, frames_list) in enumerate(zip(example_ids, annotations, frames)):
             result = {
                 "id": idx,
                 "transcription": annotation,
             }
 
+            if self.config.name != 'pre-training':
+                result["frames"] = [
+                    {"path": p, "bytes": im.read()} for p, im in frames_list
+                ]
+
             yield key, result