[Feature] Support Activation Checkpointing for ConvNeXt. (#1153)

* Support Activation Checkpointing for ConvNeXt

* Add docstring.

Co-authored-by: mzr1996 <mzr1996@163.com>

mmcls/models/backbones/convnext.py

@@ -6,6 +6,7 @@ from typing import Sequence
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
+import torch.utils.checkpoint as cp
 from mmcv.cnn.bricks import DropPath, build_activation_layer, build_norm_layer
 from mmengine.model import BaseModule, ModuleList, Sequential
 from mmengine.registry import MODELS
@@ -78,8 +79,11 @@ class ConvNeXtBlock(BaseModule):
                  mlp_ratio=4.,
                  linear_pw_conv=True,
                  drop_path_rate=0.,
-                 layer_scale_init_value=1e-6):
+                 layer_scale_init_value=1e-6,
+                 with_cp=False):
         super().__init__()
+        self.with_cp = with_cp
+
         self.depthwise_conv = nn.Conv2d(
             in_channels, in_channels, groups=in_channels, **dw_conv_cfg)
 
@@ -105,24 +109,32 @@ class ConvNeXtBlock(BaseModule):
             drop_path_rate) if drop_path_rate > 0. else nn.Identity()
 
     def forward(self, x):
-        shortcut = x
-        x = self.depthwise_conv(x)
-        x = self.norm(x)
 
-        if self.linear_pw_conv:
-            x = x.permute(0, 2, 3, 1)  # (N, C, H, W) -> (N, H, W, C)
+        def _inner_forward(x):
+            shortcut = x
+            x = self.depthwise_conv(x)
+            x = self.norm(x)
 
-        x = self.pointwise_conv1(x)
-        x = self.act(x)
-        x = self.pointwise_conv2(x)
+            if self.linear_pw_conv:
+                x = x.permute(0, 2, 3, 1)  # (N, C, H, W) -> (N, H, W, C)
 
-        if self.linear_pw_conv:
-            x = x.permute(0, 3, 1, 2)  # permute back
+            x = self.pointwise_conv1(x)
+            x = self.act(x)
+            x = self.pointwise_conv2(x)
 
-        if self.gamma is not None:
-            x = x.mul(self.gamma.view(1, -1, 1, 1))
+            if self.linear_pw_conv:
+                x = x.permute(0, 3, 1, 2)  # permute back
 
-        x = shortcut + self.drop_path(x)
+            if self.gamma is not None:
+                x = x.mul(self.gamma.view(1, -1, 1, 1))
+
+            x = shortcut + self.drop_path(x)
+            return x
+
+        if self.with_cp and x.requires_grad:
+            x = cp.checkpoint(_inner_forward, x)
+        else:
+            x = _inner_forward(x)
         return x
 
 
@@ -166,6 +178,8 @@ class ConvNeXt(BaseBackbone):
         gap_before_final_norm (bool): Whether to globally average the feature
             map before the final norm layer. In the official repo, it's only
             used in classification task. Defaults to True.
+        with_cp (bool): Use checkpoint or not. Using checkpoint will save some
+            memory while slowing down the training speed. Defaults to False.
         init_cfg (dict, optional): Initialization config dict
     """  # noqa: E501
     arch_settings = {
@@ -203,6 +217,7 @@ class ConvNeXt(BaseBackbone):
                  out_indices=-1,
                  frozen_stages=0,
                  gap_before_final_norm=True,
+                 with_cp=False,
                  init_cfg=None):
         super().__init__(init_cfg=init_cfg)
 
@@ -285,8 +300,8 @@ class ConvNeXt(BaseBackbone):
                     norm_cfg=norm_cfg,
                     act_cfg=act_cfg,
                     linear_pw_conv=linear_pw_conv,
-                    layer_scale_init_value=layer_scale_init_value)
-                for j in range(depth)
+                    layer_scale_init_value=layer_scale_init_value,
+                    with_cp=with_cp) for j in range(depth)
             ])
             block_idx += depth
 

tests/test_models/test_backbones/test_convnext.py

@@ -84,3 +84,13 @@ def test_convnext():
     for i in range(2, 4):
         assert model.downsample_layers[i].training
         assert model.stages[i].training
+
+    # Test Activation Checkpointing
+    model = ConvNeXt(arch='tiny', out_indices=-1, with_cp=True)
+    model.init_weights()
+    model.train()
+
+    imgs = torch.randn(1, 3, 224, 224)
+    feat = model(imgs)
+    assert len(feat) == 1
+    assert feat[0].shape == torch.Size([1, 768])