[Fix] Add setr & vit msg. (#635)

* [Fix] Add setr & vit msg.

* Fix init bug

* Modify init_cfg arg

* Add conv_seg init

README.md

@@ -63,6 +63,7 @@ Supported backbones:
 - [x] [ResNeSt (ArXiv'2020)](configs/resnest/README.md)
 - [x] [MobileNetV2 (CVPR'2018)](configs/mobilenet_v2/README.md)
 - [x] [MobileNetV3 (ICCV'2019)](configs/mobilenet_v3/README.md)
+- [x] [Vision Transformer (ICLR'2021)]
 
 Supported methods:
 
@@ -89,6 +90,7 @@ Supported methods:
 - [x] [DNLNet (ECCV'2020)](configs/dnlnet)
 - [x] [PointRend (CVPR'2020)](configs/point_rend)
 - [x] [CGNet (TIP'2020)](configs/cgnet)
+- [x] [SETR (CVPR'2021)](configs/setr)
 
 ## Installation
 

README_zh-CN.md

@@ -62,6 +62,7 @@ MMSegmentation 是一个基于 PyTorch 的语义分割开源工具箱。它是 O
 - [x] [ResNeSt (ArXiv'2020)](configs/resnest/README.md)
 - [x] [MobileNetV2 (CVPR'2018)](configs/mobilenet_v2/README.md)
 - [x] [MobileNetV3 (ICCV'2019)](configs/mobilenet_v3/README.md)
+- [x] [Vision Transformer (ICLR'2021)]
 
 已支持的算法：
 
@@ -87,6 +88,7 @@ MMSegmentation 是一个基于 PyTorch 的语义分割开源工具箱。它是 O
 - [x] [DNLNet (ECCV'2020)](configs/dnlnet)
 - [x] [PointRend (CVPR'2020)](configs/point_rend)
 - [x] [CGNet (TIP'2020)](configs/cgnet)
+- [x] [SETR (CVPR'2021)](configs/setr)
 
 ## 安装
 

mmseg/models/decode_heads/setr_up_head.py

@@ -1,5 +1,5 @@
 import torch.nn as nn
-from mmcv.cnn import ConvModule, build_norm_layer, constant_init
+from mmcv.cnn import ConvModule, build_norm_layer
 
 from ..builder import HEADS
 from .decode_head import BaseDecodeHead
@@ -18,6 +18,9 @@ class SETRUPHead(BaseDecodeHead):
         up_scale (int): The scale factor of interpolate. Default:4.
         kernel_size (int): The kernel size of convolution when decoding
             feature information from backbone. Default: 3.
+        init_cfg (dict | list[dict] | None): Initialization config dict.
+            Default: dict(
+                     type='Constant', val=1.0, bias=0, layer='LayerNorm').
     """
 
     def __init__(self,
@@ -25,11 +28,18 @@ class SETRUPHead(BaseDecodeHead):
                  num_convs=1,
                  up_scale=4,
                  kernel_size=3,
+                 init_cfg=[
+                     dict(type='Constant', val=1.0, bias=0, layer='LayerNorm'),
+                     dict(
+                         type='Normal',
+                         std=0.01,
+                         override=dict(name='conv_seg'))
+                 ],
                  **kwargs):
 
         assert kernel_size in [1, 3], 'kernel_size must be 1 or 3.'
 
-        super(SETRUPHead, self).__init__(**kwargs)
+        super(SETRUPHead, self).__init__(init_cfg=init_cfg, **kwargs)
 
         assert isinstance(self.in_channels, int)
 
@@ -38,7 +48,7 @@ class SETRUPHead(BaseDecodeHead):
         self.up_convs = nn.ModuleList()
         in_channels = self.in_channels
         out_channels = self.channels
-        for i in range(num_convs):
+        for _ in range(num_convs):
             self.up_convs.append(
                 nn.Sequential(
                     ConvModule(
@@ -55,12 +65,6 @@ class SETRUPHead(BaseDecodeHead):
                         align_corners=self.align_corners)))
             in_channels = out_channels
 
-    def init_weights(self):
-        for m in self.modules():
-            if isinstance(m, nn.LayerNorm):
-                constant_init(m.bias, 0)
-                constant_init(m.weight, 1.0)
-
     def forward(self, x):
         x = self._transform_inputs(x)
 

tests/test_models/test_heads/test_setr_up_head.py

@@ -16,13 +16,14 @@ def test_setr_up_head(capsys):
         # as embed_dim.
         SETRUPHead(in_channels=(32, 32), channels=16, num_classes=19)
 
-    # test init_weights of head
+    # test init_cfg of head
     head = SETRUPHead(
         in_channels=32,
         channels=16,
         norm_cfg=dict(type='SyncBN'),
-        num_classes=19)
+        num_classes=19,
-    head.init_weights()
+        init_cfg=dict(type='Kaiming'))
+    super(SETRUPHead, head).init_weights()
 
     # test inference of Naive head
     # the auxiliary head of Naive head is same as Naive head