combine subdatasets in global dataset

super-resolution-games.py

@@ -1,4 +1,5 @@
 import datasets
+from itertools import chain
 
 
 _DESCRIPTION = ''
@@ -20,13 +21,15 @@ _PROJECTS_GAME_ENGINE = [
     'GameEngine_SlayAnimationSample',
     'GameEngine_StylizedRendering',
     'GameEngine_SubwaySequencer',
-    'GameEngine_SunTemple'
+    'GameEngine_SunTemple',
+    'GameEngine_All'
 ]
 
 _PROJECTS_DOWNSCALE = [
     'Downscale_DMXPrevisSample',
     'Downscale_Dota2',
-    'Downscale_CitySample'
+    'Downscale_CitySample',
+    'Downscale_All'
 ]
 
 _PROJECTS = _PROJECTS_GAME_ENGINE + _PROJECTS_DOWNSCALE
@@ -48,13 +51,15 @@ _DESCRIPTION_DATA_GAME_ENGINE = {
     'GameEngine_SlayAnimationSample': '',
     'GameEngine_StylizedRendering': '',
     'GameEngine_SubwaySequencer': '',
-    'GameEngine_SunTemple': ''
+    'GameEngine_SunTemple': '',
+    'GameEngine_All': ''
 }
 
 _DESCRIPTION_DATA_DOWNSCALE = {
     'Downscale_DMXPrevisSample': '',
     'Downscale_Dota2': '',
-    'Downscale_CitySample': ''
+    'Downscale_CitySample': '',
+    'Downscale_All': ''
 }
 
 _DESCRIPTION_DATA = {
@@ -314,6 +319,46 @@ _DATA_FILES_GAME_ENGINE = {
     }
 }
 
+_DATA_FILES_GAME_ENGINE_KEYS = list(_DATA_FILES_GAME_ENGINE.keys())
+_DATA_FILES_GAME_ENGINE['GameEngine_All'] = {
+    'train': {
+        'r270p': sorted([
+            _DATA_FILES_GAME_ENGINE[k]['train']['r270p'] \
+                for k in _DATA_FILES_GAME_ENGINE_KEYS
+        ]),
+        'r360p': sorted([
+            _DATA_FILES_GAME_ENGINE[k]['train']['r360p'] \
+                for k in _DATA_FILES_GAME_ENGINE_KEYS
+        ]),
+        'r540p': sorted([
+            _DATA_FILES_GAME_ENGINE[k]['train']['r540p'] \
+                for k in _DATA_FILES_GAME_ENGINE_KEYS
+        ]),
+        'r1080p': sorted([
+            _DATA_FILES_GAME_ENGINE[k]['train']['r1080p'] \
+                for k in _DATA_FILES_GAME_ENGINE_KEYS
+        ])
+    },
+    'val': {
+        'r270p': sorted([
+            _DATA_FILES_GAME_ENGINE[k]['val']['r270p'] \
+                for k in _DATA_FILES_GAME_ENGINE_KEYS
+        ]),
+        'r360p': sorted([
+            _DATA_FILES_GAME_ENGINE[k]['val']['r360p'] \
+                for k in _DATA_FILES_GAME_ENGINE_KEYS
+        ]),
+        'r540p': sorted([
+            _DATA_FILES_GAME_ENGINE[k]['val']['r540p'] \
+                for k in _DATA_FILES_GAME_ENGINE_KEYS
+        ]),
+        'r1080p': sorted([
+            _DATA_FILES_GAME_ENGINE[k]['val']['r1080p'] \
+                for k in _DATA_FILES_GAME_ENGINE_KEYS
+        ])
+    }
+}
+
 _DATA_FILES_DOWNSCALE = {
     'Downscale_DMXPrevisSample': {
         'train': {
@@ -359,6 +404,46 @@ _DATA_FILES_DOWNSCALE = {
     }
 }
 
+_DATA_FILES_DOWNSCALE_KEYS = list(_DATA_FILES_DOWNSCALE.keys())
+_DATA_FILES_DOWNSCALE['Downscale_All'] = {
+    'train': {
+        'r270p': sorted([
+            _DATA_FILES_DOWNSCALE[k]['train']['r270p'] \
+                for k in _DATA_FILES_DOWNSCALE_KEYS
+        ]),
+        'r360p': sorted([
+            _DATA_FILES_DOWNSCALE[k]['train']['r360p'] \
+                for k in _DATA_FILES_DOWNSCALE_KEYS
+        ]),
+        'r540p': sorted([
+            _DATA_FILES_DOWNSCALE[k]['train']['r540p'] \
+                for k in _DATA_FILES_DOWNSCALE_KEYS
+        ]),
+        'r1080p': sorted([
+            _DATA_FILES_DOWNSCALE[k]['train']['r1080p'] \
+                for k in _DATA_FILES_DOWNSCALE_KEYS
+        ])
+    },
+    'val': {
+        'r270p': sorted([
+            _DATA_FILES_DOWNSCALE[k]['val']['r270p'] \
+                for k in _DATA_FILES_DOWNSCALE_KEYS
+        ]),
+        'r360p': sorted([
+            _DATA_FILES_DOWNSCALE[k]['val']['r360p'] \
+                for k in _DATA_FILES_DOWNSCALE_KEYS
+        ]),
+        'r540p': sorted([
+            _DATA_FILES_DOWNSCALE[k]['val']['r540p'] \
+                for k in _DATA_FILES_DOWNSCALE_KEYS
+        ]),
+        'r1080p': sorted([
+            _DATA_FILES_DOWNSCALE[k]['val']['r1080p'] \
+                for k in _DATA_FILES_DOWNSCALE_KEYS
+        ])
+    }
+}
+
 _DATA_FILES = {
     **_DATA_FILES_GAME_ENGINE,
     **_DATA_FILES_DOWNSCALE
@@ -394,30 +479,66 @@ class SuperResolutionGames(datasets.GeneratorBasedBuilder):
         data_files = self.config.data_files
         train_archives, val_archives = data_files['train'], data_files['val']
 
-        train_archives_downloaded = {
-            k: dl_manager.download(train_archives[k]) \
-                for k in train_archives.keys()
-        }
+        train_archives_downloaded = {}
+        for k in train_archives.keys():
+            train_archives_k = train_archives[k]
+            if isinstance(train_archives_k, str):
+                train_archives_downloaded[k] = \
+                    dl_manager.download(train_archives_k)
+            elif isinstance(train_archives_k, list):
+                train_archives_downloaded[k] = [
+                    dl_manager.download(train_archives_k[i]) \
+                        for i in range(len(train_archives_k))
+                ]
+            else:
+                raise TypeError()
 
-        val_archives_downloaded = {
-            k: dl_manager.download(val_archives[k]) \
-                for k in val_archives.keys()
-        }
+        val_archives_downloaded = {}
+        for k in val_archives.keys():
+            val_archives_k = val_archives[k]
+            if isinstance(val_archives_k, str):
+                val_archives_downloaded[k] = \
+                    dl_manager.download(val_archives_k)
+            elif isinstance(val_archives_k, list):
+                val_archives_downloaded[k] = [
+                    dl_manager.download(val_archives_k[i]) \
+                        for i in range(len(val_archives_k))
+                ]
+            else:
+                raise TypeError()
+
+        train_gen_kwargs = {}
+        for k in train_archives_downloaded.keys():
+            train_archives_downloaded_k = train_archives_downloaded[k]
+            if isinstance(train_archives_downloaded_k, list):
+                train_gen_kwargs[k] = chain.from_iterable([
+                    dl_manager.iter_archive(train_archives_downloaded_k[i]) \
+                        for i in range(len(train_archives_downloaded_k))
+                ])
+            else:
+                train_gen_kwargs[k] = \
+                    dl_manager.iter_archive(train_archives_downloaded_k)
+
+        val_gen_kwargs = {}
+        for k in val_archives_downloaded.keys():
+            val_archives_downloaded_k = val_archives_downloaded[k]
+            if isinstance(val_archives_downloaded_k, list):
+                val_gen_kwargs[k] = chain.from_iterable([
+                    dl_manager.iter_archive(val_archives_downloaded_k[i]) \
+                        for i in range(len(val_archives_downloaded_k))
+                ])
+            else:
+                val_gen_kwargs[k] = \
+                    dl_manager.iter_archive(val_archives_downloaded_k)
 
         splits = [
             datasets.SplitGenerator(
                 name=datasets.Split.TRAIN,
-                gen_kwargs={
-                    k: dl_manager.iter_archive(train_archives_downloaded[k]) \
-                        for k in train_archives_downloaded.keys()
-                }
+                gen_kwargs=train_gen_kwargs
             ),
             datasets.SplitGenerator(
                 name=datasets.Split.VALIDATION,
-                gen_kwargs={
-                    k: dl_manager.iter_archive(val_archives_downloaded[k]) \
-                        for k in val_archives_downloaded.keys()
-                }
+                gen_kwargs=val_gen_kwargs
             )
         ]
         return splits