From 78064ad9560f479c3aa4def2c44cb0539c5a0bdc Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Mon, 27 Jun 2022 09:43:14 +0000
Subject: [PATCH] add db++

---
 ppocr/data/imaug/operators.py          |  6 ++--
 ppocr/modeling/backbones/det_resnet.py |  2 +-
 ppocr/modeling/necks/db_fpn.py         | 47 ++++++++++++++++----------
 3 files changed, 33 insertions(+), 22 deletions(-)

diff --git a/ppocr/data/imaug/operators.py b/ppocr/data/imaug/operators.py
index a83dc648..ee5492de 100644
--- a/ppocr/data/imaug/operators.py
+++ b/ppocr/data/imaug/operators.py
@@ -242,8 +242,8 @@ class DetResizeForTest(object):
         if 'image_shape' in kwargs:
             self.image_shape = kwargs['image_shape']
             self.resize_type = 1
-            if 'keep_ratio' in kwargs:  ######
-                self.keep_ratio = kwargs['keep_ratio']  #######
+            if 'keep_ratio' in kwargs:
+                self.keep_ratio = kwargs['keep_ratio']
         elif 'limit_side_len' in kwargs:
             self.limit_side_len = kwargs['limit_side_len']
             self.limit_type = kwargs.get('limit_type', 'min')
@@ -273,7 +273,7 @@ class DetResizeForTest(object):
     def resize_image_type1(self, img):
         resize_h, resize_w = self.image_shape
         ori_h, ori_w = img.shape[:2]  # (h, w, c)
-        if self.keep_ratio:  ########
+        if self.keep_ratio:
             resize_w = ori_w * resize_h / ori_h
             N = math.ceil(resize_w / 32)
             resize_w = N * 32
diff --git a/ppocr/modeling/backbones/det_resnet.py b/ppocr/modeling/backbones/det_resnet.py
index 4c75b663..3d8ce445 100644
--- a/ppocr/modeling/backbones/det_resnet.py
+++ b/ppocr/modeling/backbones/det_resnet.py
@@ -1,4 +1,4 @@
-# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/ppocr/modeling/necks/db_fpn.py b/ppocr/modeling/necks/db_fpn.py
index b46d7c46..8c3f52a3 100644
--- a/ppocr/modeling/necks/db_fpn.py
+++ b/ppocr/modeling/necks/db_fpn.py
@@ -105,7 +105,7 @@ class DSConv(nn.Layer):
 
 
 class DBFPN(nn.Layer):
-    def __init__(self, in_channels, out_channels, use_asf=None, **kwargs):
+    def __init__(self, in_channels, out_channels, use_asf=False, **kwargs):
         super(DBFPN, self).__init__()
         self.out_channels = out_channels
         self.use_asf = use_asf
@@ -164,7 +164,7 @@ class DBFPN(nn.Layer):
             weight_attr=ParamAttr(initializer=weight_attr),
             bias_attr=False)
 
-        if self.use_asf:
+        if self.use_asf is True:
             self.asf = ASFBlock(self.out_channels, self.out_channels // 4)
 
     def forward(self, x):
@@ -192,7 +192,7 @@ class DBFPN(nn.Layer):
 
         fuse = paddle.concat([p5, p4, p3, p2], axis=1)
 
-        if self.use_asf:
+        if self.use_asf is True:
             fuse = self.asf(fuse, [p5, p4, p3, p2])
 
         return fuse
@@ -367,7 +367,19 @@ class LKPAN(nn.Layer):
 
 
 class ASFBlock(nn.Layer):
+    """
+    This code is refered from:
+        https://github.com/MhLiao/DB/blob/master/decoders/feature_attention.py
+    """
+
     def __init__(self, in_channels, inter_channels, out_features_num=4):
+        """
+        Adaptive Scale Fusion (ASF) block of DBNet++
+        Args:
+            in_channels: the number of channels in the input data
+            inter_channels: the number of middle channels
+            out_features_num: the number of fused stages
+        """
         super(ASFBlock, self).__init__()
         weight_attr = paddle.nn.initializer.KaimingUniform()
         self.in_channels = in_channels
@@ -375,39 +387,38 @@ class ASFBlock(nn.Layer):
         self.out_features_num = out_features_num
         self.conv = nn.Conv2D(in_channels, inter_channels, 3, padding=1)
 
-        self.attention_block_1 = nn.Sequential(
+        self.spatial_scale = nn.Sequential(
             #Nx1xHxW
             nn.Conv2D(
-                1,
-                1,
-                3,
+                in_channels=1,
+                out_channels=1,
+                kernel_size=3,
                 bias_attr=False,
                 padding=1,
                 weight_attr=ParamAttr(initializer=weight_attr)),
             nn.ReLU(),
             nn.Conv2D(
-                1,
-                1,
-                1,
+                in_channels=1,
+                out_channels=1,
+                kernel_size=1,
                 bias_attr=False,
                 weight_attr=ParamAttr(initializer=weight_attr)),
             nn.Sigmoid())
 
-        self.attention_block_2 = nn.Sequential(
+        self.channel_scale = nn.Sequential(
             nn.Conv2D(
-                inter_channels,
-                out_features_num,
-                1,
+                in_channels=inter_channels,
+                out_channels=out_features_num,
+                kernel_size=1,
                 bias_attr=False,
                 weight_attr=ParamAttr(initializer=weight_attr)),
             nn.Sigmoid())
 
     def forward(self, fuse_features, features_list):
         fuse_features = self.conv(fuse_features)
-        attention_scores = self.attention_block_1(
-            paddle.mean(
-                fuse_features, axis=1, keepdim=True)) + fuse_features
-        attention_scores = self.attention_block_2(attention_scores)
+        spatial_x = paddle.mean(fuse_features, axis=1, keepdim=True)
+        attention_scores = self.spatial_scale(spatial_x) + fuse_features
+        attention_scores = self.channel_scale(attention_scores)
         assert len(features_list) == self.out_features_num
 
         out_list = []
-- 
GitLab