update

coco_captions.py

@@ -38,7 +38,8 @@ class COCOCaptionsConfig(datasets.BuilderConfig):
         url = 'http://images.cocodataset.org/zips/'
         self.train_image_url = url + 'train2017.zip'
         self.val_image_url = url + 'val2017.zip'
-        self.annotation_urls = annotation_urls
+        self.train_annotation_urls = annotation_urls['train']
+        self.val_annotation_urls = annotation_urls['validation']
 
 
 class COCOCaptions(datasets.GeneratorBasedBuilder):
@@ -47,8 +48,8 @@ class COCOCaptions(datasets.GeneratorBasedBuilder):
             description=_DESCRIPTION,
             homepage=_HOMEPAGE,
             annotation_urls={
-                'train': 'data/captions_train.jsonl',
-                'validation': 'data/captions_validation.jsonl'
+                'train': 'data/captions_train.zip',
+                'validation': 'data/captions_validation.zip'
             },
         )
     ]
@@ -73,25 +74,28 @@ class COCOCaptions(datasets.GeneratorBasedBuilder):
         train_image_path = dl_manager.download_and_extract(
             self.config.train_image_url
         )
-        validation_image_path = dl_manager.download_and_extract(
+        val_image_path = dl_manager.download_and_extract(
             self.config.val_image_url
         )
-        annotation_paths = dl_manager.download_and_extract(
-            self.config.annotation_urls
+        train_annotation_paths = dl_manager.download_and_extract(
+            self.config.train_annotation_urls
+        )
+        val_annotation_paths = dl_manager.download_and_extract(
+            self.config.val_annotation_urls
         )
         return [
             datasets.SplitGenerator(
                 name=datasets.Split.TRAIN,
                 gen_kwargs={
                     'image_path': f'{train_image_path}/train2017',
-                    'annotation_path': annotation_paths['train']
+                    'annotation_path': f'{train_annotation_paths}/captions_train.jsonl'
                 }
             ),
             datasets.SplitGenerator(
                 name=datasets.Split.VALIDATION,
                 gen_kwargs={
-                    'image_path': f'{validation_image_path}/val2017',
-                    'annotation_path': annotation_paths['validation']
+                    'image_path': f'{val_image_path}/val2017',
+                    'annotation_path': f'{val_annotation_paths}/captions_validation.jsonl'
                 }
             )
         ]