Merge branch 'dygraph' of https://github.com/PaddlePaddle/PaddleOCR into test_v10

ppocr/modeling/heads/rec_sar_head.py

+# copyright (c) 2021 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.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from: 
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textrecog/encoders/sar_encoder.py
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textrecog/decoders/sar_decoder.py
+"""
+
 from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function

ppocr/modeling/necks/fpn.py

@@ -11,16 +11,31 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+"""
+This code is refer from:
+https://github.com/whai362/PSENet/blob/python3/models/neck/fpn.py
+"""
 
 import paddle.nn as nn
 import paddle
 import math
 import paddle.nn.functional as F
 
+
 class Conv_BN_ReLU(nn.Layer):
-    def __init__(self, in_planes, out_planes, kernel_size=1, stride=1, padding=0):
+    def __init__(self,
+                 in_planes,
+                 out_planes,
+                 kernel_size=1,
+                 stride=1,
+                 padding=0):
         super(Conv_BN_ReLU, self).__init__()
-        self.conv = nn.Conv2D(in_planes, out_planes, kernel_size=kernel_size, stride=stride, padding=padding,
+        self.conv = nn.Conv2D(
+            in_planes,
+            out_planes,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=padding,
             bias_attr=False)
         self.bn = nn.BatchNorm2D(out_planes, momentum=0.1)
         self.relu = nn.ReLU()
@@ -28,46 +43,69 @@ class Conv_BN_ReLU(nn.Layer):
         for m in self.sublayers():
             if isinstance(m, nn.Conv2D):
                 n = m._kernel_size[0] * m._kernel_size[1] * m._out_channels
-                m.weight = paddle.create_parameter(shape=m.weight.shape, dtype='float32', default_initializer=paddle.nn.initializer.Normal(0, math.sqrt(2. / n)))
+                m.weight = paddle.create_parameter(
+                    shape=m.weight.shape,
+                    dtype='float32',
+                    default_initializer=paddle.nn.initializer.Normal(
+                        0, math.sqrt(2. / n)))
             elif isinstance(m, nn.BatchNorm2D):
-                m.weight = paddle.create_parameter(shape=m.weight.shape, dtype='float32', default_initializer=paddle.nn.initializer.Constant(1.0))
-                m.bias = paddle.create_parameter(shape=m.bias.shape, dtype='float32', default_initializer=paddle.nn.initializer.Constant(0.0))
+                m.weight = paddle.create_parameter(
+                    shape=m.weight.shape,
+                    dtype='float32',
+                    default_initializer=paddle.nn.initializer.Constant(1.0))
+                m.bias = paddle.create_parameter(
+                    shape=m.bias.shape,
+                    dtype='float32',
+                    default_initializer=paddle.nn.initializer.Constant(0.0))
 
     def forward(self, x):
         return self.relu(self.bn(self.conv(x)))
 
+
 class FPN(nn.Layer):
     def __init__(self, in_channels, out_channels):
         super(FPN, self).__init__()
 
         # Top layer
-        self.toplayer_ = Conv_BN_ReLU(in_channels[3], out_channels, kernel_size=1, stride=1, padding=0)
+        self.toplayer_ = Conv_BN_ReLU(
+            in_channels[3], out_channels, kernel_size=1, stride=1, padding=0)
         # Lateral layers
-        self.latlayer1_ = Conv_BN_ReLU(in_channels[2], out_channels, kernel_size=1, stride=1, padding=0)
+        self.latlayer1_ = Conv_BN_ReLU(
+            in_channels[2], out_channels, kernel_size=1, stride=1, padding=0)
 
-        self.latlayer2_ = Conv_BN_ReLU(in_channels[1], out_channels, kernel_size=1, stride=1, padding=0)
+        self.latlayer2_ = Conv_BN_ReLU(
+            in_channels[1], out_channels, kernel_size=1, stride=1, padding=0)
 
-        self.latlayer3_ = Conv_BN_ReLU(in_channels[0], out_channels, kernel_size=1, stride=1, padding=0)
+        self.latlayer3_ = Conv_BN_ReLU(
+            in_channels[0], out_channels, kernel_size=1, stride=1, padding=0)
 
         # Smooth layers
-        self.smooth1_ = Conv_BN_ReLU(out_channels, out_channels, kernel_size=3, stride=1, padding=1)
-
-        self.smooth2_ = Conv_BN_ReLU(out_channels, out_channels, kernel_size=3, stride=1, padding=1)
+        self.smooth1_ = Conv_BN_ReLU(
+            out_channels, out_channels, kernel_size=3, stride=1, padding=1)
 
-        self.smooth3_ = Conv_BN_ReLU(out_channels, out_channels, kernel_size=3, stride=1, padding=1)
+        self.smooth2_ = Conv_BN_ReLU(
+            out_channels, out_channels, kernel_size=3, stride=1, padding=1)
 
+        self.smooth3_ = Conv_BN_ReLU(
+            out_channels, out_channels, kernel_size=3, stride=1, padding=1)
 
         self.out_channels = out_channels * 4
         for m in self.sublayers():
             if isinstance(m, nn.Conv2D):
                 n = m._kernel_size[0] * m._kernel_size[1] * m._out_channels
-                m.weight = paddle.create_parameter(shape=m.weight.shape, dtype='float32',
-                                                   default_initializer=paddle.nn.initializer.Normal(0,
-                                                                                                    math.sqrt(2. / n)))
+                m.weight = paddle.create_parameter(
+                    shape=m.weight.shape,
+                    dtype='float32',
+                    default_initializer=paddle.nn.initializer.Normal(
+                        0, math.sqrt(2. / n)))
             elif isinstance(m, nn.BatchNorm2D):
-                m.weight = paddle.create_parameter(shape=m.weight.shape, dtype='float32',
+                m.weight = paddle.create_parameter(
+                    shape=m.weight.shape,
+                    dtype='float32',
                     default_initializer=paddle.nn.initializer.Constant(1.0))
-                m.bias = paddle.create_parameter(shape=m.bias.shape, dtype='float32',
+                m.bias = paddle.create_parameter(
+                    shape=m.bias.shape,
+                    dtype='float32',
                     default_initializer=paddle.nn.initializer.Constant(0.0))
 
     def _upsample(self, x, scale=1):
@@ -81,15 +119,15 @@ class FPN(nn.Layer):
         p5 = self.toplayer_(f5)
 
         f4 = self.latlayer1_(f4)
-        p4 = self._upsample_add(p5, f4,2)
+        p4 = self._upsample_add(p5, f4, 2)
         p4 = self.smooth1_(p4)
 
         f3 = self.latlayer2_(f3)
-        p3 = self._upsample_add(p4, f3,2)
+        p3 = self._upsample_add(p4, f3, 2)
         p3 = self.smooth2_(p3)
 
         f2 = self.latlayer3_(f2)
-        p2 = self._upsample_add(p3, f2,2)
+        p2 = self._upsample_add(p3, f2, 2)
         p2 = self.smooth3_(p2)
 
         p3 = self._upsample(p3, 2)

ppocr/modeling/transforms/stn.py

@@ -11,7 +11,10 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-
+"""
+This code is refer from:
+https://github.com/ayumiymk/aster.pytorch/blob/master/lib/models/stn_head.py
+"""
 from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function

ppocr/modeling/transforms/tps.py

@@ -11,6 +11,10 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+"""
+This code is refer from:
+https://github.com/clovaai/deep-text-recognition-benchmark/blob/master/modules/transformation.py
+"""
 
 from __future__ import absolute_import
 from __future__ import division

ppocr/modeling/transforms/tps_spatial_transformer.py

@@ -11,6 +11,10 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+"""
+This code is refer from:
+https://github.com/ayumiymk/aster.pytorch/blob/master/lib/models/tps_spatial_transformer.py
+"""
 from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function

ppocr/postprocess/locality_aware_nms.py

 """
 Locality aware nms.
+This code is refered from: https://github.com/songdejia/EAST/blob/master/locality_aware_nms.py
 """
 
 import numpy as np

ppocr/postprocess/pse_postprocess/pse/README.md

 ## 编译
-code from https://github.com/whai362/pan_pp.pytorch
+This code is refer from:
+https://github.com/whai362/PSENet/blob/python3/models/post_processing/pse
 ```python
 python3 setup.py build_ext --inplace
 ```

ppocr/postprocess/pse_postprocess/pse_postprocess.py

-# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+# copyright (c) 2021 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.
@@ -11,6 +11,10 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+"""
+This code is refer from:
+https://github.com/whai362/PSENet/blob/python3/models/head/psenet_head.py
+"""
 
 from __future__ import absolute_import
 from __future__ import division
@@ -47,7 +51,8 @@ class PSEPostProcess(object):
         pred = outs_dict['maps']
         if not isinstance(pred, paddle.Tensor):
             pred = paddle.to_tensor(pred)
-        pred = F.interpolate(pred, scale_factor=4 // self.scale, mode='bilinear')
+        pred = F.interpolate(
+            pred, scale_factor=4 // self.scale, mode='bilinear')
 
         score = F.sigmoid(pred[:, 0, :, :])
 
@@ -60,7 +65,9 @@ class PSEPostProcess(object):
 
         boxes_batch = []
         for batch_index in range(pred.shape[0]):
-            boxes, scores = self.boxes_from_bitmap(score[batch_index], kernels[batch_index], shape_list[batch_index])
+            boxes, scores = self.boxes_from_bitmap(score[batch_index],
+                                                   kernels[batch_index],
+                                                   shape_list[batch_index])
 
             boxes_batch.append({'points': boxes, 'scores': scores})
         return boxes_batch
@@ -98,15 +105,14 @@ class PSEPostProcess(object):
                 mask = np.zeros((box_height, box_width), np.uint8)
                 mask[points[:, 1], points[:, 0]] = 255
 
-                contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+                contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL,
+                                               cv2.CHAIN_APPROX_SIMPLE)
                 bbox = np.squeeze(contours[0], 1)
             else:
                 raise NotImplementedError
 
-            bbox[:, 0] = np.clip(
-                np.round(bbox[:, 0] / ratio_w), 0, src_w)
-            bbox[:, 1] = np.clip(
-                np.round(bbox[:, 1] / ratio_h), 0, src_h)
+            bbox[:, 0] = np.clip(np.round(bbox[:, 0] / ratio_w), 0, src_w)
+            bbox[:, 1] = np.clip(np.round(bbox[:, 1] / ratio_h), 0, src_h)
             boxes.append(bbox)
             scores.append(score_i)
         return boxes, scores

ppocr/utils/iou.py

-# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+# copyright (c) 2021 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.
@@ -11,18 +11,23 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+"""
+This code is refer from:
+https://github.com/whai362/PSENet/blob/python3/models/loss/iou.py
+"""
 
 import paddle
 
 EPS = 1e-6
 
+
 def iou_single(a, b, mask, n_class):
     valid = mask == 1
     a = a.masked_select(valid)
     b = b.masked_select(valid)
     miou = []
     for i in range(n_class):
-        if a.shape == [0] and a.shape==b.shape:
+        if a.shape == [0] and a.shape == b.shape:
             inter = paddle.to_tensor(0.0)
             union = paddle.to_tensor(0.0)
         else:
@@ -32,6 +37,7 @@ def iou_single(a, b, mask, n_class):
     miou = sum(miou) / len(miou)
     return miou
 
+
 def iou(a, b, mask, n_class=2, reduce=True):
     batch_size = a.shape[0]
 
@@ -39,7 +45,7 @@ def iou(a, b, mask, n_class=2, reduce=True):
     b = b.reshape([batch_size, -1])
     mask = mask.reshape([batch_size, -1])
 
-    iou = paddle.zeros((batch_size,), dtype='float32')
+    iou = paddle.zeros((batch_size, ), dtype='float32')
     for i in range(batch_size):
         iou[i] = iou_single(a[i], b[i], mask[i], n_class)
 

ppocr/utils/logging.py

-# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+# copyright (c) 2020 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.
@@ -11,6 +11,10 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+"""
+This code is refer from:
+https://github.com/WenmuZhou/PytorchOCR/blob/master/torchocr/utils/logging.py
+"""
 
 import os
 import sys