[Refactor] Delete redundant set_random_seed function (open-mmlab#104)

* refactor set_random_seed

* add unittests

* fix unittests error

* fix lint

* avoid bc breaking

mmrazor/apis/__init__.py

@@ -2,3 +2,4 @@
 from .mmcls import *  # noqa: F401,F403
 from .mmdet import *  # noqa: F401,F403
 from .mmseg import *  # noqa: F401,F403
+from .utils import set_random_seed  # noqa: F401

mmrazor/apis/mmcls/__init__.py

@@ -1,5 +1,5 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 from .inference import init_mmcls_model
-from .train import set_random_seed, train_model
+from .train import set_random_seed, train_mmcls_model
 
-__all__ = ['set_random_seed', 'train_model', 'init_mmcls_model']
+__all__ = ['train_mmcls_model', 'init_mmcls_model', 'set_random_seed']

mmrazor/apis/mmcls/train.py

@@ -20,7 +20,8 @@
 
 
 def set_random_seed(seed, deterministic=False):
-    """Set random seed.
+    """Import `set_random_seed` function here was deprecated in v0.3 and will
+    be removed in v0.5.
 
     Args:
         seed (int): Seed to be used.
@@ -29,6 +30,11 @@ def set_random_seed(seed, deterministic=False):
             to True and ``torch.backends.cudnn.benchmark`` to False.
             Default: False.
     """
+    warnings.warn(
+        'Deprecated in v0.3 and will be removed in v0.5, '
+        'please import `set_random_seed` directly from `mmrazor.apis`',
+        category=DeprecationWarning)
+
     random.seed(seed)
     np.random.seed(seed)
     torch.manual_seed(seed)
@@ -38,14 +44,14 @@ def set_random_seed(seed, deterministic=False):
         torch.backends.cudnn.benchmark = False
 
 
-def train_model(model,
-                dataset,
-                cfg,
-                distributed=False,
-                validate=False,
-                timestamp=None,
-                device='cuda',
-                meta=None):
+def train_mmcls_model(model,
+                      dataset,
+                      cfg,
+                      distributed=False,
+                      validate=False,
+                      timestamp=None,
+                      device='cuda',
+                      meta=None):
     """Copy from mmclassification and modify some codes.
 
     This is an ugly implementation, and will be deprecated in the future. In

mmrazor/apis/mmdet/__init__.py

@@ -6,6 +6,6 @@
 
 if mmdet is not None:
     from .inference import init_mmdet_model
-    from .train import set_random_seed, train_detector
+    from .train import set_random_seed, train_mmdet_model
 
-    __all__ = ['set_random_seed', 'train_detector', 'init_mmdet_model']
+    __all__ = ['train_mmdet_model', 'init_mmdet_model', 'set_random_seed']

mmrazor/apis/mmdet/train.py

@@ -19,7 +19,8 @@
 
 
 def set_random_seed(seed, deterministic=False):
-    """Set random seed.
+    """Import `set_random_seed` function here was deprecated in v0.3 and will
+    be removed in v0.5.
 
     Args:
         seed (int): Seed to be used.
@@ -28,6 +29,11 @@ def set_random_seed(seed, deterministic=False):
             to True and ``torch.backends.cudnn.benchmark`` to False.
             Default: False.
     """
+    warnings.warn(
+        'Deprecated in v0.3 and will be removed in v0.5, '
+        'please import `set_random_seed` directly from `mmrazor.apis`',
+        category=DeprecationWarning)
+
     random.seed(seed)
     np.random.seed(seed)
     torch.manual_seed(seed)
@@ -37,13 +43,13 @@ def set_random_seed(seed, deterministic=False):
         torch.backends.cudnn.benchmark = False
 
 
-def train_detector(model,
-                   dataset,
-                   cfg,
-                   distributed=False,
-                   validate=False,
-                   timestamp=None,
-                   meta=None):
+def train_mmdet_model(model,
+                      dataset,
+                      cfg,
+                      distributed=False,
+                      validate=False,
+                      timestamp=None,
+                      meta=None):
     """Copy from mmdetection and modify some codes.
 
     This is an ugly implementation, and will be deprecated in the future. In

mmrazor/apis/mmseg/__init__.py

@@ -6,6 +6,6 @@
 
 if mmseg:
     from .inference import init_mmseg_model
-    from .train import set_random_seed, train_segmentor
+    from .train import set_random_seed, train_mmseg_model
 
-    __all__ = ['set_random_seed', 'train_segmentor', 'init_mmseg_model']
+    __all__ = ['train_mmseg_model', 'init_mmseg_model', 'set_random_seed']

mmrazor/apis/mmseg/train.py

@@ -16,7 +16,8 @@
 
 
 def set_random_seed(seed, deterministic=False):
-    """Set random seed.
+    """Import `set_random_seed` function here was deprecated in v0.3 and will
+    be removed in v0.5.
 
     Args:
         seed (int): Seed to be used.
@@ -25,6 +26,11 @@ def set_random_seed(seed, deterministic=False):
             to True and ``torch.backends.cudnn.benchmark`` to False.
             Default: False.
     """
+    warnings.warn(
+        'Deprecated in v0.3 and will be removed in v0.5, '
+        'please import `set_random_seed` directly from `mmrazor.apis`',
+        category=DeprecationWarning)
+
     random.seed(seed)
     np.random.seed(seed)
     torch.manual_seed(seed)
@@ -34,13 +40,13 @@ def set_random_seed(seed, deterministic=False):
         torch.backends.cudnn.benchmark = False
 
 
-def train_segmentor(model,
-                    dataset,
-                    cfg,
-                    distributed=False,
-                    validate=False,
-                    timestamp=None,
-                    meta=None):
+def train_mmseg_model(model,
+                      dataset,
+                      cfg,
+                      distributed=False,
+                      validate=False,
+                      timestamp=None,
+                      meta=None):
     """Copy from mmsegmentation and modify some codes.
 
     This is an ugly implementation, and will be deprecated in the future. In

mmrazor/apis/utils.py

@@ -0,0 +1,24 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+import random
+
+import numpy as np
+import torch
+
+
+def set_random_seed(seed: int, deterministic: bool = False) -> None:
+    """Set random seed.
+
+    Args:
+        seed (int): Seed to be used.
+        deterministic (bool): Whether to set the deterministic option for
+            CUDNN backend, i.e., set ``torch.backends.cudnn.deterministic``
+            to True and ``torch.backends.cudnn.benchmark`` to False.
+            Default: False.
+    """
+    random.seed(seed)
+    np.random.seed(seed)
+    torch.manual_seed(seed)
+    torch.cuda.manual_seed_all(seed)
+    if deterministic:
+        torch.backends.cudnn.deterministic = True
+        torch.backends.cudnn.benchmark = False

tests/deprecated_api/test_remove_0-5.py

@@ -0,0 +1,18 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+import pytest
+
+
+def test_v0_5_deprecated_set_random_seed() -> None:
+    warn_msg = 'Deprecated in v0.3 and will be removed in v0.5, ' \
+        'please import `set_random_seed` directly from `mmrazor.apis`'
+    from mmrazor.apis.mmcls import set_random_seed
+    with pytest.deprecated_call(match=warn_msg):
+        set_random_seed(123)
+
+    from mmrazor.apis.mmdet import set_random_seed
+    with pytest.deprecated_call(match=warn_msg):
+        set_random_seed(123)
+
+    from mmrazor.apis.mmseg import set_random_seed
+    with pytest.deprecated_call(match=warn_msg):
+        set_random_seed(123)

tests/test_apis/test_inference.py

@@ -9,11 +9,26 @@
 from mmrazor.apis import init_mmcls_model, init_mmdet_model, init_mmseg_model
 
 
-def test_init_mmcls_model():
+def _sync_bn2bn(config: mmcv.Config) -> None:
+
+    def dfs(cfg_dict) -> None:
+        if isinstance(cfg_dict, dict):
+            for k, v in cfg_dict.items():
+                if k == 'norm_cfg':
+                    if v['type'] == 'SyncBN':
+                        v['type'] = 'BN'
+                dfs(v)
+
+    dfs(config._cfg_dict)
+
+
+def test_init_mmcls_model() -> None:
     from mmcls.datasets import ImageNet
 
     config_file = 'configs/nas/spos/spos_subnet_shufflenetv2_8xb128_in1k.py'
     config = mmcv.Config.fromfile(config_file)
+    # Replace SyncBN with BN to inference on CPU
+    _sync_bn2bn(config)
 
     mutable_file = 'configs/nas/spos/SPOS_SHUFFLENETV2_330M_IN1k_PAPER.yaml'
     model = init_mmcls_model(
@@ -34,10 +49,12 @@ def test_init_mmcls_model():
     assert result.get('pred_class') is not None
 
 
-def test_init_mmdet_model():
+def test_init_mmdet_model() -> None:
     config_file = \
         'configs/nas/detnas/detnas_subnet_frcnn_shufflenetv2_fpn_1x_coco.py'
     config = mmcv.Config.fromfile(config_file)
+    # Replace SyncBN with BN to inference on CPU
+    _sync_bn2bn(config)
 
     mutable_file = \
         'configs/nas/detnas/DETNAS_FRCNN_SHUFFLENETV2_340M_COCO_MMRAZOR.yaml'
@@ -52,17 +69,12 @@ def test_init_mmdet_model():
     assert isinstance(result, list)
 
 
-def test_init_mmseg_model():
+def test_init_mmseg_model() -> None:
     config_file = 'configs/distill/cwd/' \
         'cwd_cls_head_pspnet_r101_d8_pspnet_r18_d8_512x1024_cityscapes_80k.py'
     config = mmcv.Config.fromfile(config_file)
-
     # Replace SyncBN with BN to inference on CPU
-    norm_cfg = dict(type='BN', requires_grad=True)
-    model_config = config.algorithm.architecture
-    model_config.model.backbone.norm_cfg = norm_cfg
-    model_config.model.decode_head.norm_cfg = norm_cfg
-    model_config.model.auxiliary_head.norm_cfg = norm_cfg
+    _sync_bn2bn(config)
 
     # Enable test time augmentation
     config.data.test.pipeline[1].flip = True

tests/test_apis/test_utils.py

@@ -0,0 +1,20 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+import numpy as np
+import torch
+
+from mmrazor.apis import set_random_seed
+
+
+def test_set_random_seed() -> None:
+    set_random_seed(123, False)
+    x1 = torch.rand(3, 3)
+    x2 = np.random.rand(3, 3)
+
+    set_random_seed(123, True)
+    assert torch.backends.cudnn.deterministic
+    assert not torch.backends.cudnn.benchmark
+    y1 = torch.rand(3, 3)
+    y2 = np.random.rand(3, 3)
+
+    assert torch.allclose(x1, y1, 1e-6)
+    assert np.allclose(x2, y2, 1e-6)

tools/mmcls/train_mmcls.py

@@ -15,7 +15,7 @@
 from mmcv.runner import get_dist_info, init_dist
 
 # Differences from mmclassification
-from mmrazor.apis.mmcls.train import set_random_seed, train_model
+from mmrazor.apis import set_random_seed, train_mmcls_model
 from mmrazor.models import build_algorithm
 from mmrazor.utils import setup_multi_processes
 
@@ -179,7 +179,7 @@ def main():
             config=cfg.pretty_text,
             CLASSES=datasets[0].CLASSES)
     # add an attribute for visualization convenience
-    train_model(
+    train_mmcls_model(
         # Difference from mmclassification
         # replace `model` to `algorithm`
         algorithm,

tools/mmdet/train_mmdet.py

@@ -23,7 +23,7 @@
 from mmdet.datasets import build_dataset
 from mmdet.utils import collect_env, get_root_logger
 
-from mmrazor.apis import set_random_seed, train_detector
+from mmrazor.apis import set_random_seed, train_mmdet_model
 from mmrazor.models import build_algorithm
 from mmrazor.utils import setup_multi_processes
 
@@ -190,7 +190,7 @@ def main():
             CLASSES=datasets[0].CLASSES)
     # add an attribute for visualization convenience
     algorithm.CLASSES = datasets[0].CLASSES
-    train_detector(
+    train_mmdet_model(
         algorithm,
         datasets,
         cfg,

tools/mmseg/train_mmseg.py

@@ -24,7 +24,7 @@
 from mmseg.utils import collect_env, get_root_logger
 
 # Differences from mmdetection
-from mmrazor.apis.mmseg import set_random_seed, train_segmentor
+from mmrazor.apis import set_random_seed, train_mmseg_model
 from mmrazor.models.builder import build_algorithm
 from mmrazor.utils import setup_multi_processes
 
@@ -210,7 +210,7 @@ def main():
     algorithm.CLASSES = datasets[0].CLASSES
     # passing checkpoint meta for saving best checkpoint
     meta.update(cfg.checkpoint_config.meta)
-    train_segmentor(
+    train_mmseg_model(
         # Difference from mmsegmentation
         # replace `model` to `algorithm`
         algorithm,