set test_mode for mmdet (#920)

* fix

* update

mmdeploy/codebase/base/task.py

@@ -68,7 +68,7 @@ class BaseTask(metaclass=ABCMeta):
     def build_dataset(self,
                       dataset_cfg: Union[str, mmcv.Config],
                       dataset_type: str = 'val',
-                      is_sort_dataset: bool = True,
+                      is_sort_dataset: bool = False,
                       **kwargs) -> Dataset:
         """Build dataset for different codebase.
 
@@ -80,6 +80,7 @@ class BaseTask(metaclass=ABCMeta):
             is_sort_dataset (bool): When 'True', the dataset will be sorted
                 by image shape in ascending order if 'dataset_cfg'
                 contains information about height and width.
+                Default is `False`.
 
         Returns:
             Dataset: The built dataset.

mmdeploy/codebase/mmdet/deploy/mmdetection.py

@@ -62,9 +62,19 @@ class MMDetection(MMCodebase):
 
         data_cfg = dataset_cfg.data[dataset_type]
         samples_per_gpu = dataset_cfg.data.get('samples_per_gpu', 1)
-        if samples_per_gpu > 1:
-            # Replace 'ImageToTensor' to 'DefaultFormatBundle'
-            data_cfg.pipeline = replace_ImageToTensor(data_cfg.pipeline)
+        if isinstance(data_cfg, dict):
+            data_cfg.test_mode = True
+            if samples_per_gpu > 1:
+                # Replace 'ImageToTensor' to 'DefaultFormatBundle'
+                data_cfg.pipeline = replace_ImageToTensor(data_cfg.pipeline)
+        elif isinstance(data_cfg, list):
+            for ds_cfg in data_cfg:
+                ds_cfg.test_mode = True
+                if samples_per_gpu > 1:
+                    ds_cfg.pipeline = replace_ImageToTensor(ds_cfg.pipeline)
+        else:
+            raise TypeError(f'Unsupported type {type(data_cfg)}')
+
         dataset = build_dataset_mmdet(data_cfg)
 
         return dataset