yujunjun/PaddleOCR
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/PaddleOCR.git synced 2025-10-31 17:59:11 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #4551 from WenmuZhou/copyright

Browse Source 
add refer for some code
This commit is contained in:
MissPenguin 2021-11-04 19:41:06 +08:00 committed by GitHub
commit 5aa14c5f11
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
14 changed files with 209 additions and 67 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
ppocr/data/imaug/iaa_augment.py
View File

@ -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

18
ppocr/data/imaug/make_border_map.py
View File

@ -1,4 +1,20 @@
# -*- 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

49
ppocr/data/imaug/make_pse_gt.py
View File

@ -1,4 +1,16 @@
# -*- 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:

18
ppocr/data/imaug/make_shrink_map.py
View File

@ -1,4 +1,20 @@
# -*- 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

18
ppocr/data/imaug/random_crop_data.py
View File

@ -1,4 +1,20 @@
# -*- 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

4
ppocr/data/imaug/text_image_aug/augment.py
View File

@ -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

4
ppocr/data/imaug/text_image_aug/warp_mls.py
View File

@ -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

4
ppocr/losses/det_pse_loss.py
View File

@ -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

20
ppocr/modeling/heads/det_pse_head.py
View File

@ -1,4 +1,4 @@
# 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)

80
ppocr/modeling/necks/fpn.py
View File

@ -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.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.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)

3
ppocr/postprocess/pse_postprocess/pse/README.md
View File

@ -1,5 +1,6 @@
## 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
```

22
ppocr/postprocess/pse_postprocess/pse_postprocess.py
View File

@ -1,4 +1,4 @@
# 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

12
ppocr/utils/iou.py
View File

@ -1,4 +1,4 @@
# 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)

6
ppocr/utils/logging.py
View File

@ -1,4 +1,4 @@
# 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