Merge branch 'dygraph' of https://github.com/PaddlePaddle/PaddleOCR into paddle2onnx

ppocr/data/imaug/east_process.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 refered from: 
+https://github.com/songdejia/EAST/blob/master/data_utils.py
+"""
 import math
 import cv2
 import numpy as np
@@ -24,10 +27,10 @@ __all__ = ['EASTProcessTrain']
 
 class EASTProcessTrain(object):
     def __init__(self,
-                 image_shape = [512, 512],
-                 background_ratio = 0.125,
-                 min_crop_side_ratio = 0.1,
-                 min_text_size = 10,
+                 image_shape=[512, 512],
+                 background_ratio=0.125,
+                 min_crop_side_ratio=0.1,
+                 min_text_size=10,
                  **kwargs):
         self.input_size = image_shape[1]
         self.random_scale = np.array([0.5, 1, 2.0, 3.0])
@@ -282,12 +285,7 @@ class EASTProcessTrain(object):
                 1.0 / max(min(poly_h, poly_w), 1.0)
         return score_map, geo_map, training_mask
 
-    def crop_area(self,
-                  im,
-                  polys,
-                  tags,
-                  crop_background=False,
-                  max_tries=50):
+    def crop_area(self, im, polys, tags, crop_background=False, max_tries=50):
         """
         make random crop from the input image
         :param im:
@@ -435,5 +433,4 @@ class EASTProcessTrain(object):
         data['score_map'] = score_map
         data['geo_map'] = geo_map
         data['training_mask'] = training_mask
-        # print(im.shape, score_map.shape, geo_map.shape, training_mask.shape)
-        return data
+        return data
\ No newline at end of file

ppocr/data/imaug/iaa_augment.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/WenmuZhou/DBNet.pytorch/blob/master/data_loader/modules/iaa_augment.py
+"""
 from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function

ppocr/data/imaug/make_border_map.py

-# -*- coding:utf-8 -*- 
+# 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.
+# 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/WenmuZhou/DBNet.pytorch/blob/master/data_loader/modules/make_border_map.py
+"""
 
 from __future__ import absolute_import
 from __future__ import division

ppocr/data/imaug/make_pse_gt.py

-# -*- coding:utf-8 -*- 
+# 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.
 
 from __future__ import absolute_import
 from __future__ import division
@@ -12,12 +24,8 @@ from shapely.geometry import Polygon
 
 __all__ = ['MakePseGt']
 
-class MakePseGt(object):
-    r'''
-    Making binary mask from detection data with ICDAR format.
-    Typically following the process of class `MakeICDARData`.
-    '''
 
+class MakePseGt(object):
     def __init__(self, kernel_num=7, size=640, min_shrink_ratio=0.4, **kwargs):
         self.kernel_num = kernel_num
         self.min_shrink_ratio = min_shrink_ratio
@@ -38,16 +46,20 @@ class MakePseGt(object):
             text_polys *= scale
 
         gt_kernels = []
-        for i in range(1,self.kernel_num+1):
+        for i in range(1, self.kernel_num + 1):
             # s1->sn, from big to small
-            rate = 1.0 - (1.0 - self.min_shrink_ratio) / (self.kernel_num - 1) * i
-            text_kernel, ignore_tags = self.generate_kernel(image.shape[0:2], rate, text_polys, ignore_tags)
+            rate = 1.0 - (1.0 - self.min_shrink_ratio) / (self.kernel_num - 1
+                                                          ) * i
+            text_kernel, ignore_tags = self.generate_kernel(
+                image.shape[0:2], rate, text_polys, ignore_tags)
             gt_kernels.append(text_kernel)
 
         training_mask = np.ones(image.shape[0:2], dtype='uint8')
         for i in range(text_polys.shape[0]):
             if ignore_tags[i]:
-                cv2.fillPoly(training_mask, text_polys[i].astype(np.int32)[np.newaxis, :, :], 0)
+                cv2.fillPoly(training_mask,
+                             text_polys[i].astype(np.int32)[np.newaxis, :, :],
+                             0)
 
         gt_kernels = np.array(gt_kernels)
         gt_kernels[gt_kernels > 0] = 1
@@ -59,16 +71,25 @@ class MakePseGt(object):
         data['mask'] = training_mask.astype('float32')
         return data
 
-    def generate_kernel(self, img_size, shrink_ratio, text_polys, ignore_tags=None):
+    def generate_kernel(self,
+                        img_size,
+                        shrink_ratio,
+                        text_polys,
+                        ignore_tags=None):
+        """
+        Refer to part of the code:
+        https://github.com/open-mmlab/mmocr/blob/main/mmocr/datasets/pipelines/textdet_targets/base_textdet_targets.py
+        """
+
         h, w = img_size
         text_kernel = np.zeros((h, w), dtype=np.float32)
         for i, poly in enumerate(text_polys):
             polygon = Polygon(poly)
-            distance = polygon.area * (1 - shrink_ratio * shrink_ratio) / (polygon.length + 1e-6)
+            distance = polygon.area * (1 - shrink_ratio * shrink_ratio) / (
+                polygon.length + 1e-6)
             subject = [tuple(l) for l in poly]
             pco = pyclipper.PyclipperOffset()
-            pco.AddPath(subject, pyclipper.JT_ROUND,
-                        pyclipper.ET_CLOSEDPOLYGON)
+            pco.AddPath(subject, pyclipper.JT_ROUND, pyclipper.ET_CLOSEDPOLYGON)
             shrinked = np.array(pco.Execute(-distance))
 
             if len(shrinked) == 0 or shrinked.size == 0:

ppocr/data/imaug/make_shrink_map.py

-# -*- coding:utf-8 -*- 
+# 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.
+# 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/WenmuZhou/DBNet.pytorch/blob/master/data_loader/modules/make_shrink_map.py
+"""
 
 from __future__ import absolute_import
 from __future__ import division

ppocr/data/imaug/random_crop_data.py

-# -*- coding:utf-8 -*- 
+# 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.
+# 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/WenmuZhou/DBNet.pytorch/blob/master/data_loader/modules/random_crop_data.py
+"""
 
 from __future__ import absolute_import
 from __future__ import division

ppocr/data/imaug/sast_process.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 part code is refered from: 
+https://github.com/songdejia/EAST/blob/master/data_utils.py
+"""
 import math
 import cv2
 import numpy as np

ppocr/data/imaug/text_image_aug/augment.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/RubanSeven/Text-Image-Augmentation-python/blob/master/augment.py
+"""
 
 import numpy as np
 from .warp_mls import WarpMLS

ppocr/data/imaug/text_image_aug/warp_mls.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/RubanSeven/Text-Image-Augmentation-python/blob/master/warp_mls.py
+"""
 
 import numpy as np
 
@@ -161,4 +165,4 @@ class WarpMLS:
         dst = np.clip(dst, 0, 255)
         dst = np.array(dst, dtype=np.uint8)
 
-        return dst
+        return dst
\ No newline at end of file

ppocr/losses/ace_loss.py

@@ -11,6 +11,9 @@
 # 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/viig99/LS-ACELoss
+
 from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
@@ -32,7 +35,7 @@ class ACELoss(nn.Layer):
     def __call__(self, predicts, batch):
         if isinstance(predicts, (list, tuple)):
             predicts = predicts[-1]
-            
+
         B, N = predicts.shape[:2]
         div = paddle.to_tensor([N]).astype('float32')
 

ppocr/losses/center_loss.py

@@ -12,6 +12,8 @@
 #See the License for the specific language governing permissions and
 #limitations under the License.
 
+# This code is refer from: https://github.com/KaiyangZhou/pytorch-center-loss
+
 from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
@@ -27,6 +29,7 @@ class CenterLoss(nn.Layer):
     """
     Reference: Wen et al. A Discriminative Feature Learning Approach for Deep Face Recognition. ECCV 2016.
     """
+
     def __init__(self,
                  num_classes=6625,
                  feat_dim=96,

ppocr/losses/det_basic_loss.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/WenmuZhou/DBNet.pytorch/blob/master/models/losses/basic_loss.py
+"""
 from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
@@ -147,4 +150,4 @@ class BCELoss(nn.Layer):
 
     def forward(self, input, label, mask=None, weight=None, name=None):
         loss = F.binary_cross_entropy(input, label, reduction=self.reduction)
-        return loss
+        return loss
\ No newline at end of file

ppocr/losses/det_db_loss.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/WenmuZhou/DBNet.pytorch/blob/master/models/losses/DB_loss.py
+"""
 
 from __future__ import absolute_import
 from __future__ import division

ppocr/losses/det_pse_loss.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/whai362/PSENet/blob/python3/models/head/psenet_head.py
+"""
 
 import paddle
 from paddle import nn

ppocr/losses/rec_nrtr_loss.py

@@ -22,7 +22,7 @@ class NRTRLoss(nn.Layer):
             log_prb = F.log_softmax(pred, axis=1)
             non_pad_mask = paddle.not_equal(
                 tgt, paddle.zeros(
-                    tgt.shape, dtype='int64'))
+                    tgt.shape, dtype=tgt.dtype))
             loss = -(one_hot * log_prb).sum(axis=1)
             loss = loss.masked_select(non_pad_mask).mean()
         else:

ppocr/modeling/backbones/rec_mv1_enhance.py

@@ -12,6 +12,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+# This code is refer from: https://github.com/PaddlePaddle/PaddleClas/blob/develop/ppcls/arch/backbone/legendary_models/pp_lcnet.py
+
 from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function

ppocr/modeling/backbones/rec_resnet_31.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/layers/conv_layer.py
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textrecog/backbones/resnet31_ocr.py
+"""
+
 from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
@@ -18,12 +37,12 @@ def conv3x3(in_channel, out_channel, stride=1):
         kernel_size=3,
         stride=stride,
         padding=1,
-        bias_attr=False
-    )
+        bias_attr=False)
 
 
 class BasicBlock(nn.Layer):
     expansion = 1
+
     def __init__(self, in_channels, channels, stride=1, downsample=False):
         super().__init__()
         self.conv1 = conv3x3(in_channels, channels, stride)
@@ -34,9 +53,13 @@ class BasicBlock(nn.Layer):
         self.downsample = downsample
         if downsample:
             self.downsample = nn.Sequential(
-                nn.Conv2D(in_channels, channels * self.expansion, 1, stride, bias_attr=False),
-                nn.BatchNorm2D(channels * self.expansion),
-            )
+                nn.Conv2D(
+                    in_channels,
+                    channels * self.expansion,
+                    1,
+                    stride,
+                    bias_attr=False),
+                nn.BatchNorm2D(channels * self.expansion), )
         else:
             self.downsample = nn.Sequential()
         self.stride = stride
@@ -57,7 +80,7 @@ class BasicBlock(nn.Layer):
         out += residual
         out = self.relu(out)
 
-        return out        
+        return out
 
 
 class ResNet31(nn.Layer):
@@ -69,12 +92,13 @@ class ResNet31(nn.Layer):
         out_indices (None | Sequence[int]): Indices of output stages.
         last_stage_pool (bool): If True, add `MaxPool2d` layer to last stage.
     '''
-    def __init__(self, 
-                in_channels=3, 
-                layers=[1, 2, 5, 3],
-                channels=[64, 128, 256, 256, 512, 512, 512],
-                out_indices=None,
-                last_stage_pool=False):
+
+    def __init__(self,
+                 in_channels=3,
+                 layers=[1, 2, 5, 3],
+                 channels=[64, 128, 256, 256, 512, 512, 512],
+                 out_indices=None,
+                 last_stage_pool=False):
         super(ResNet31, self).__init__()
         assert isinstance(in_channels, int)
         assert isinstance(last_stage_pool, bool)
@@ -83,46 +107,56 @@ class ResNet31(nn.Layer):
         self.last_stage_pool = last_stage_pool
 
         # conv 1 (Conv Conv)
-        self.conv1_1 = nn.Conv2D(in_channels, channels[0], kernel_size=3, stride=1, padding=1)
+        self.conv1_1 = nn.Conv2D(
+            in_channels, channels[0], kernel_size=3, stride=1, padding=1)
         self.bn1_1 = nn.BatchNorm2D(channels[0])
         self.relu1_1 = nn.ReLU()
 
-        self.conv1_2 = nn.Conv2D(channels[0], channels[1], kernel_size=3, stride=1, padding=1)
+        self.conv1_2 = nn.Conv2D(
+            channels[0], channels[1], kernel_size=3, stride=1, padding=1)
         self.bn1_2 = nn.BatchNorm2D(channels[1])
         self.relu1_2 = nn.ReLU()
 
         # conv 2 (Max-pooling, Residual block, Conv)
-        self.pool2 = nn.MaxPool2D(kernel_size=2, stride=2, padding=0, ceil_mode=True)
+        self.pool2 = nn.MaxPool2D(
+            kernel_size=2, stride=2, padding=0, ceil_mode=True)
         self.block2 = self._make_layer(channels[1], channels[2], layers[0])
-        self.conv2 = nn.Conv2D(channels[2], channels[2], kernel_size=3, stride=1, padding=1)
+        self.conv2 = nn.Conv2D(
+            channels[2], channels[2], kernel_size=3, stride=1, padding=1)
         self.bn2 = nn.BatchNorm2D(channels[2])
         self.relu2 = nn.ReLU()
 
         # conv 3 (Max-pooling, Residual block, Conv)
-        self.pool3 = nn.MaxPool2D(kernel_size=2, stride=2, padding=0, ceil_mode=True)
+        self.pool3 = nn.MaxPool2D(
+            kernel_size=2, stride=2, padding=0, ceil_mode=True)
         self.block3 = self._make_layer(channels[2], channels[3], layers[1])
-        self.conv3 = nn.Conv2D(channels[3], channels[3], kernel_size=3, stride=1, padding=1)
+        self.conv3 = nn.Conv2D(
+            channels[3], channels[3], kernel_size=3, stride=1, padding=1)
         self.bn3 = nn.BatchNorm2D(channels[3])
         self.relu3 = nn.ReLU()
 
         # conv 4 (Max-pooling, Residual block, Conv)
-        self.pool4 = nn.MaxPool2D(kernel_size=(2, 1), stride=(2, 1), padding=0, ceil_mode=True)
+        self.pool4 = nn.MaxPool2D(
+            kernel_size=(2, 1), stride=(2, 1), padding=0, ceil_mode=True)
         self.block4 = self._make_layer(channels[3], channels[4], layers[2])
-        self.conv4 = nn.Conv2D(channels[4], channels[4], kernel_size=3, stride=1, padding=1)
+        self.conv4 = nn.Conv2D(
+            channels[4], channels[4], kernel_size=3, stride=1, padding=1)
         self.bn4 = nn.BatchNorm2D(channels[4])
         self.relu4 = nn.ReLU()
 
         # conv 5 ((Max-pooling), Residual block, Conv)
         self.pool5 = None
         if self.last_stage_pool:
-            self.pool5 = nn.MaxPool2D(kernel_size=2, stride=2, padding=0, ceil_mode=True)
+            self.pool5 = nn.MaxPool2D(
+                kernel_size=2, stride=2, padding=0, ceil_mode=True)
         self.block5 = self._make_layer(channels[4], channels[5], layers[3])
-        self.conv5 = nn.Conv2D(channels[5], channels[5], kernel_size=3, stride=1, padding=1)
+        self.conv5 = nn.Conv2D(
+            channels[5], channels[5], kernel_size=3, stride=1, padding=1)
         self.bn5 = nn.BatchNorm2D(channels[5])
         self.relu5 = nn.ReLU()
 
         self.out_channels = channels[-1]
-    
+
     def _make_layer(self, input_channels, output_channels, blocks):
         layers = []
         for _ in range(blocks):
@@ -130,19 +164,19 @@ class ResNet31(nn.Layer):
             if input_channels != output_channels:
                 downsample = nn.Sequential(
                     nn.Conv2D(
-                        input_channels, 
-                        output_channels, 
-                        kernel_size=1, 
-                        stride=1, 
+                        input_channels,
+                        output_channels,
+                        kernel_size=1,
+                        stride=1,
                         bias_attr=False),
-                    nn.BatchNorm2D(output_channels),
-                )
-                
-            layers.append(BasicBlock(input_channels, output_channels, downsample=downsample))
+                    nn.BatchNorm2D(output_channels), )
+
+            layers.append(
+                BasicBlock(
+                    input_channels, output_channels, downsample=downsample))
             input_channels = output_channels
         return nn.Sequential(*layers)
 
-
     def forward(self, x):
         x = self.conv1_1(x)
         x = self.bn1_1(x)
@@ -166,11 +200,11 @@ class ResNet31(nn.Layer):
             x = block_layer(x)
             x = conv_layer(x)
             x = bn_layer(x)
-            x= relu_layer(x)
+            x = relu_layer(x)
 
             outs.append(x)
-        
+
         if self.out_indices is not None:
             return tuple([outs[i] for i in self.out_indices])
-        
+
         return x

ppocr/modeling/backbones/rec_resnet_aster.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/resnet_aster.py
+"""
 import paddle
 import paddle.nn as nn
 

ppocr/modeling/heads/det_pse_head.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.
@@ -11,22 +11,24 @@
 # 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 paddle import nn
 
 
 class PSEHead(nn.Layer):
-    def __init__(self,
-                 in_channels,
-                 hidden_dim=256,
-                 out_channels=7,
-                 **kwargs):
+    def __init__(self, in_channels, hidden_dim=256, out_channels=7, **kwargs):
         super(PSEHead, self).__init__()
-        self.conv1 = nn.Conv2D(in_channels, hidden_dim, kernel_size=3, stride=1, padding=1)
+        self.conv1 = nn.Conv2D(
+            in_channels, hidden_dim, kernel_size=3, stride=1, padding=1)
         self.bn1 = nn.BatchNorm2D(hidden_dim)
         self.relu1 = nn.ReLU()
 
-        self.conv2 = nn.Conv2D(hidden_dim, out_channels, kernel_size=1, stride=1, padding=0)
-
+        self.conv2 = nn.Conv2D(
+            hidden_dim, out_channels, kernel_size=1, stride=1, padding=0)
 
     def forward(self, x, **kwargs):
         out = self.conv1(x)