Ema (#421)

* add ema hook

* add ema hook resume

* add ema hook test

* fix typo

* fix according to comment

* delete logger

* fix according to comment

* fix unitest

* fix typo

* fix according to comment

* change to resume_from

* typo

* fix isort

mmcv/runner/__init__.py

@@ -6,8 +6,8 @@ from .dist_utils import get_dist_info, init_dist, master_only
 from .epoch_based_runner import EpochBasedRunner, Runner
 from .fp16_utils import auto_fp16, force_fp32
 from .hooks import (HOOKS, CheckpointHook, ClosureHook, DistSamplerSeedHook,
-                    Fp16OptimizerHook, Hook, IterTimerHook, LoggerHook,
-                    LrUpdaterHook, MlflowLoggerHook, OptimizerHook,
+                    EMAHook, Fp16OptimizerHook, Hook, IterTimerHook,
+                    LoggerHook, LrUpdaterHook, MlflowLoggerHook, OptimizerHook,
                     PaviLoggerHook, SyncBuffersHook, TensorboardLoggerHook,
                     TextLoggerHook, WandbLoggerHook)
 from .iter_based_runner import IterBasedRunner, IterLoader
@@ -30,5 +30,5 @@ __all__ = [
     'OPTIMIZER_BUILDERS', 'OPTIMIZERS', 'DefaultOptimizerConstructor',
     'build_optimizer', 'build_optimizer_constructor', 'IterLoader',
     'set_random_seed', 'auto_fp16', 'force_fp32', 'Fp16OptimizerHook',
-    'SyncBuffersHook'
+    'SyncBuffersHook', 'EMAHook'
 ]

mmcv/runner/base_runner.py

@@ -307,10 +307,13 @@ class BaseRunner(metaclass=ABCMeta):
                resume_optimizer=True,
                map_location='default'):
         if map_location == 'default':
-            device_id = torch.cuda.current_device()
-            checkpoint = self.load_checkpoint(
-                checkpoint,
-                map_location=lambda storage, loc: storage.cuda(device_id))
+            if torch.cuda.is_available():
+                device_id = torch.cuda.current_device()
+                checkpoint = self.load_checkpoint(
+                    checkpoint,
+                    map_location=lambda storage, loc: storage.cuda(device_id))
+            else:
+                checkpoint = self.load_checkpoint(checkpoint)
         else:
             checkpoint = self.load_checkpoint(
                 checkpoint, map_location=map_location)

mmcv/runner/hooks/__init__.py

@@ -1,6 +1,7 @@
 # Copyright (c) Open-MMLab. All rights reserved.
 from .checkpoint import CheckpointHook
 from .closure import ClosureHook
+from .ema import EMAHook
 from .hook import HOOKS, Hook
 from .iter_timer import IterTimerHook
 from .logger import (LoggerHook, MlflowLoggerHook, PaviLoggerHook,
@@ -17,5 +18,5 @@ __all__ = [
     'OptimizerHook', 'Fp16OptimizerHook', 'IterTimerHook',
     'DistSamplerSeedHook', 'EmptyCacheHook', 'LoggerHook', 'MlflowLoggerHook',
     'PaviLoggerHook', 'TextLoggerHook', 'TensorboardLoggerHook',
-    'WandbLoggerHook', 'MomentumUpdaterHook', 'SyncBuffersHook'
+    'WandbLoggerHook', 'MomentumUpdaterHook', 'SyncBuffersHook', 'EMAHook'
 ]

mmcv/runner/hooks/ema.py

@@ -0,0 +1,88 @@
+from ...parallel import is_module_wrapper
+from ..hooks.hook import HOOKS, Hook
+
+
+@HOOKS.register_module()
+class EMAHook(Hook):
+    r"""Exponential Moving Average Hook.
+
+    Use Exponential Moving Average on all parameters of model in training
+    process. All parameters have a ema backup, which update by the formula
+    as below. EMAHook takes priority over EvalHook and CheckpointSaverHook.
+
+        .. math::
+
+            \text{Xema_{t+1}} = (1 - \text{momentum}) \times
+            \text{Xema_{t}} +  \text{momentum} \times X_t
+
+    Args:
+        momentum (float): The momentum used for updating ema parameter.
+            Defaults to 0.0002.
+        interval (int): Update ema parameter every interval iteration.
+            Defaults to 1.
+        warm_up (int): During first warm_up steps, we may use smaller momentum
+            to update ema parameters more slowly. Defaults to 100.
+        resume_from (str): The checkpoint path. Defaults to None.
+    """
+
+    def __init__(self,
+                 momentum=0.0002,
+                 interval=1,
+                 warm_up=100,
+                 resume_from=None):
+        assert isinstance(interval, int) and interval > 0
+        self.warm_up = warm_up
+        self.interval = interval
+        assert momentum > 0 and momentum < 1
+        self.momentum = momentum**interval
+        self.checkpoint = resume_from
+
+    def before_run(self, runner):
+        """To resume model with it's ema parameters more friendly.
+
+        Register ema parameter as ``named_buffer`` to model
+        """
+        model = runner.model
+        if is_module_wrapper(model):
+            model = model.module
+        self.param_ema_buffer = {}
+        self.model_parameters = dict(model.named_parameters(recurse=True))
+        for name, value in self.model_parameters.items():
+            # "." is not allowed in module's buffer name
+            buffer_name = f"ema_{name.replace('.', '_')}"
+            self.param_ema_buffer[name] = buffer_name
+            model.register_buffer(buffer_name, value.data.clone())
+        self.model_buffers = dict(model.named_buffers(recurse=True))
+        if self.checkpoint is not None:
+            runner.resume(self.checkpoint)
+
+    def after_train_iter(self, runner):
+        """Update ema parameter every self.interval iterations."""
+        curr_step = runner.iter
+        # We warm up the momentum considering the instability at beginning
+        momentum = min(self.momentum,
+                       (1 + curr_step) / (self.warm_up + curr_step))
+        if curr_step % self.interval != 0:
+            return
+        for name, parameter in self.model_parameters.items():
+            buffer_name = self.param_ema_buffer[name]
+            buffer_parameter = self.model_buffers[buffer_name]
+            buffer_parameter.mul_(1 - momentum).add_(momentum, parameter.data)
+
+    def after_train_epoch(self, runner):
+        """We load parameter values from ema backup to model before the
+        EvalHook."""
+        self._swap_ema_parameters()
+
+    def before_train_epoch(self, runner):
+        """We recover model's parameter from ema backup after last epoch's
+        EvalHook."""
+        self._swap_ema_parameters()
+
+    def _swap_ema_parameters(self):
+        """Swap the parameter of model with parameter in ema_buffer."""
+        for name, value in self.model_parameters.items():
+            temp = value.data.clone()
+            ema_buffer = self.model_buffers[self.param_ema_buffer[name]]
+            value.data.copy_(ema_buffer.data)
+            ema_buffer.data.copy_(temp)

tests/test_runner/test_hooks.py

@@ -14,13 +14,85 @@ from unittest.mock import MagicMock, call
 import pytest
 import torch
 import torch.nn as nn
+from torch.nn.init import constant_
 from torch.utils.data import DataLoader
 
-from mmcv.runner import (EpochBasedRunner, IterTimerHook, MlflowLoggerHook,
-                         PaviLoggerHook, WandbLoggerHook)
+from mmcv.runner import (CheckpointHook, EMAHook, EpochBasedRunner,
+                         IterTimerHook, MlflowLoggerHook, PaviLoggerHook,
+                         WandbLoggerHook)
 from mmcv.runner.hooks.lr_updater import CosineRestartLrUpdaterHook
 
 
+def test_ema_hook():
+    """xdoctest -m tests/test_hooks.py test_ema_hook."""
+
+    class DemoModel(nn.Module):
+
+        def __init__(self):
+            super().__init__()
+            self.conv = nn.Conv2d(
+                in_channels=1,
+                out_channels=2,
+                kernel_size=1,
+                padding=1,
+                bias=True)
+            self._init_weight()
+
+        def _init_weight(self):
+            constant_(self.conv.weight, 0)
+            constant_(self.conv.bias, 0)
+
+        def forward(self, x):
+            return self.conv(x).sum()
+
+        def train_step(self, x, optimizer, **kwargs):
+            return dict(loss=self(x))
+
+        def val_step(self, x, optimizer, **kwargs):
+            return dict(loss=self(x))
+
+    loader = DataLoader(torch.ones((1, 1, 1, 1)))
+    runner = _build_demo_runner()
+    demo_model = DemoModel()
+    runner.model = demo_model
+    emahook = EMAHook(momentum=0.1, interval=2, warm_up=100, resume_from=None)
+    checkpointhook = CheckpointHook(interval=1, by_epoch=True)
+    runner.register_hook(emahook, priority='HIGHEST')
+    runner.register_hook(checkpointhook)
+    runner.run([loader, loader], [('train', 1), ('val', 1)], 1)
+    checkpoint = torch.load(f'{runner.work_dir}/epoch_1.pth')
+    contain_ema_buffer = False
+    for name, value in checkpoint['state_dict'].items():
+        if 'ema' in name:
+            contain_ema_buffer = True
+            assert value.sum() == 0
+            value.fill_(1)
+        else:
+            assert value.sum() == 0
+    assert contain_ema_buffer
+    torch.save(checkpoint, f'{runner.work_dir}/epoch_1.pth')
+    work_dir = runner.work_dir
+    resume_ema_hook = EMAHook(
+        momentum=0.5, warm_up=0, resume_from=f'{work_dir}/epoch_1.pth')
+    runner = _build_demo_runner()
+    runner.model = demo_model
+    runner.register_hook(resume_ema_hook, priority='HIGHEST')
+    checkpointhook = CheckpointHook(interval=1, by_epoch=True)
+    runner.register_hook(checkpointhook)
+    runner.run([loader, loader], [('train', 1), ('val', 1)], 2)
+    checkpoint = torch.load(f'{runner.work_dir}/epoch_2.pth')
+    contain_ema_buffer = False
+    for name, value in checkpoint['state_dict'].items():
+        if 'ema' in name:
+            contain_ema_buffer = True
+            assert value.sum() == 2
+        else:
+            assert value.sum() == 1
+    assert contain_ema_buffer
+    shutil.rmtree(runner.work_dir)
+    shutil.rmtree(work_dir)
+
+
 def test_pavi_hook():
     sys.modules['pavi'] = MagicMock()