Merge branch 'dygraph' into bm_dyg

ppocr/losses/__init__.py

@@ -13,28 +13,39 @@
 # limitations under the License.
 
 import copy
+import paddle
+import paddle.nn as nn
 
+# det loss
+from .det_db_loss import DBLoss
+from .det_east_loss import EASTLoss
+from .det_sast_loss import SASTLoss
 
-def build_loss(config):
-    # det loss
-    from .det_db_loss import DBLoss
-    from .det_east_loss import EASTLoss
-    from .det_sast_loss import SASTLoss
+# rec loss
+from .rec_ctc_loss import CTCLoss
+from .rec_att_loss import AttentionLoss
+from .rec_srn_loss import SRNLoss
+
+# cls loss
+from .cls_loss import ClsLoss
+
+# e2e loss
+from .e2e_pg_loss import PGLoss
 
-    # rec loss
-    from .rec_ctc_loss import CTCLoss
-    from .rec_att_loss import AttentionLoss
-    from .rec_srn_loss import SRNLoss
+# basic loss function
+from .basic_loss import DistanceLoss
 
-    # cls loss
-    from .cls_loss import ClsLoss
+# combined loss function
+from .combined_loss import CombinedLoss
 
-    # e2e loss
-    from .e2e_pg_loss import PGLoss
+# table loss
+from .table_att_loss import TableAttentionLoss
+
+def build_loss(config):
     support_dict = [
         'DBLoss', 'EASTLoss', 'SASTLoss', 'CTCLoss', 'ClsLoss', 'AttentionLoss',
-        'SRNLoss', 'PGLoss']
-
+        'SRNLoss', 'PGLoss', 'CombinedLoss', 'TableAttentionLoss'
+    ]
     config = copy.deepcopy(config)
     module_name = config.pop('name')
     assert module_name in support_dict, Exception('loss only support {}'.format(

ppocr/losses/basic_loss.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.
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+from paddle.nn import L1Loss
+from paddle.nn import MSELoss as L2Loss
+from paddle.nn import SmoothL1Loss
+
+
+class CELoss(nn.Layer):
+    def __init__(self, epsilon=None):
+        super().__init__()
+        if epsilon is not None and (epsilon <= 0 or epsilon >= 1):
+            epsilon = None
+        self.epsilon = epsilon
+
+    def _labelsmoothing(self, target, class_num):
+        if target.shape[-1] != class_num:
+            one_hot_target = F.one_hot(target, class_num)
+        else:
+            one_hot_target = target
+        soft_target = F.label_smooth(one_hot_target, epsilon=self.epsilon)
+        soft_target = paddle.reshape(soft_target, shape=[-1, class_num])
+        return soft_target
+
+    def forward(self, x, label):
+        loss_dict = {}
+        if self.epsilon is not None:
+            class_num = x.shape[-1]
+            label = self._labelsmoothing(label, class_num)
+            x = -F.log_softmax(x, axis=-1)
+            loss = paddle.sum(x * label, axis=-1)
+        else:
+            if label.shape[-1] == x.shape[-1]:
+                label = F.softmax(label, axis=-1)
+                soft_label = True
+            else:
+                soft_label = False
+            loss = F.cross_entropy(x, label=label, soft_label=soft_label)
+        return loss
+
+
+class DMLLoss(nn.Layer):
+    """
+    DMLLoss
+    """
+
+    def __init__(self, act=None):
+        super().__init__()
+        if act is not None:
+            assert act in ["softmax", "sigmoid"]
+        if act == "softmax":
+            self.act = nn.Softmax(axis=-1)
+        elif act == "sigmoid":
+            self.act = nn.Sigmoid()
+        else:
+            self.act = None
+
+    def forward(self, out1, out2):
+        if self.act is not None:
+            out1 = self.act(out1)
+            out2 = self.act(out2)
+
+        log_out1 = paddle.log(out1)
+        log_out2 = paddle.log(out2)
+        loss = (F.kl_div(
+            log_out1, out2, reduction='batchmean') + F.kl_div(
+                log_out2, out1, reduction='batchmean')) / 2.0
+        return loss
+
+
+class DistanceLoss(nn.Layer):
+    """
+    DistanceLoss:
+        mode: loss mode
+    """
+
+    def __init__(self, mode="l2", **kargs):
+        super().__init__()
+        assert mode in ["l1", "l2", "smooth_l1"]
+        if mode == "l1":
+            self.loss_func = nn.L1Loss(**kargs)
+        elif mode == "l2":
+            self.loss_func = nn.MSELoss(**kargs)
+        elif mode == "smooth_l1":
+            self.loss_func = nn.SmoothL1Loss(**kargs)
+
+    def forward(self, x, y):
+        return self.loss_func(x, y)

ppocr/losses/cls_loss.py

@@ -24,7 +24,7 @@ class ClsLoss(nn.Layer):
         super(ClsLoss, self).__init__()
         self.loss_func = nn.CrossEntropyLoss(reduction='mean')
 
-    def __call__(self, predicts, batch):
+    def forward(self, predicts, batch):
         label = batch[1]
         loss = self.loss_func(input=predicts, label=label)
         return {'loss': loss}

ppocr/losses/combined_loss.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# 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.
+
+import paddle
+import paddle.nn as nn
+
+from .distillation_loss import DistillationCTCLoss
+from .distillation_loss import DistillationDMLLoss
+from .distillation_loss import DistillationDistanceLoss
+
+
+class CombinedLoss(nn.Layer):
+    """
+    CombinedLoss:
+        a combionation of loss function
+    """
+
+    def __init__(self, loss_config_list=None):
+        super().__init__()
+        self.loss_func = []
+        self.loss_weight = []
+        assert isinstance(loss_config_list, list), (
+            'operator config should be a list')
+        for config in loss_config_list:
+            assert isinstance(config,
+                              dict) and len(config) == 1, "yaml format error"
+            name = list(config)[0]
+            param = config[name]
+            assert "weight" in param, "weight must be in param, but param just contains {}".format(
+                param.keys())
+            self.loss_weight.append(param.pop("weight"))
+            self.loss_func.append(eval(name)(**param))
+
+    def forward(self, input, batch, **kargs):
+        loss_dict = {}
+        for idx, loss_func in enumerate(self.loss_func):
+            loss = loss_func(input, batch, **kargs)
+            if isinstance(loss, paddle.Tensor):
+                loss = {"loss_{}_{}".format(str(loss), idx): loss}
+            weight = self.loss_weight[idx]
+            loss = {
+                "{}_{}".format(key, idx): loss[key] * weight
+                for key in loss
+            }
+            loss_dict.update(loss)
+        loss_dict["loss"] = paddle.add_n(list(loss_dict.values()))
+        return loss_dict

ppocr/losses/distillation_loss.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.
+
+import paddle
+import paddle.nn as nn
+
+from .rec_ctc_loss import CTCLoss
+from .basic_loss import DMLLoss
+from .basic_loss import DistanceLoss
+
+
+class DistillationDMLLoss(DMLLoss):
+    """
+    """
+
+    def __init__(self, model_name_pairs=[], act=None, key=None,
+                 name="loss_dml"):
+        super().__init__(act=act)
+        assert isinstance(model_name_pairs, list)
+        self.key = key
+        self.model_name_pairs = model_name_pairs
+        self.name = name
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+        for idx, pair in enumerate(self.model_name_pairs):
+            out1 = predicts[pair[0]]
+            out2 = predicts[pair[1]]
+            if self.key is not None:
+                out1 = out1[self.key]
+                out2 = out2[self.key]
+            loss = super().forward(out1, out2)
+            if isinstance(loss, dict):
+                for key in loss:
+                    loss_dict["{}_{}_{}_{}".format(key, pair[0], pair[1],
+                                                   idx)] = loss[key]
+            else:
+                loss_dict["{}_{}".format(self.name, idx)] = loss
+        return loss_dict
+
+
+class DistillationCTCLoss(CTCLoss):
+    def __init__(self, model_name_list=[], key=None, name="loss_ctc"):
+        super().__init__()
+        self.model_name_list = model_name_list
+        self.key = key
+        self.name = name
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+        for idx, model_name in enumerate(self.model_name_list):
+            out = predicts[model_name]
+            if self.key is not None:
+                out = out[self.key]
+            loss = super().forward(out, batch)
+            if isinstance(loss, dict):
+                for key in loss:
+                    loss_dict["{}_{}_{}".format(self.name, model_name,
+                                                idx)] = loss[key]
+            else:
+                loss_dict["{}_{}".format(self.name, model_name)] = loss
+        return loss_dict
+
+
+class DistillationDistanceLoss(DistanceLoss):
+    """
+    """
+
+    def __init__(self,
+                 mode="l2",
+                 model_name_pairs=[],
+                 key=None,
+                 name="loss_distance",
+                 **kargs):
+        super().__init__(mode=mode, **kargs)
+        assert isinstance(model_name_pairs, list)
+        self.key = key
+        self.model_name_pairs = model_name_pairs
+        self.name = name + "_l2"
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+        for idx, pair in enumerate(self.model_name_pairs):
+            out1 = predicts[pair[0]]
+            out2 = predicts[pair[1]]
+            if self.key is not None:
+                out1 = out1[self.key]
+                out2 = out2[self.key]
+            loss = super().forward(out1, out2)
+            if isinstance(loss, dict):
+                for key in loss:
+                    loss_dict["{}_{}_{}".format(self.name, key, idx)] = loss[
+                        key]
+            else:
+                loss_dict["{}_{}_{}_{}".format(self.name, pair[0], pair[1],
+                                               idx)] = loss
+        return loss_dict

ppocr/losses/rec_ctc_loss.py

@@ -25,7 +25,7 @@ class CTCLoss(nn.Layer):
         super(CTCLoss, self).__init__()
         self.loss_func = nn.CTCLoss(blank=0, reduction='none')
 
-    def __call__(self, predicts, batch):
+    def forward(self, predicts, batch):
         predicts = predicts.transpose((1, 0, 2))
         N, B, _ = predicts.shape
         preds_lengths = paddle.to_tensor([N] * B, dtype='int64')

ppocr/losses/table_att_loss.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.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+from paddle.nn import functional as F
+from paddle import fluid
+
+class TableAttentionLoss(nn.Layer):
+    def __init__(self, structure_weight, loc_weight, use_giou=False, giou_weight=1.0, **kwargs):
+        super(TableAttentionLoss, self).__init__()
+        self.loss_func = nn.CrossEntropyLoss(weight=None, reduction='none')
+        self.structure_weight = structure_weight
+        self.loc_weight = loc_weight
+        self.use_giou = use_giou
+        self.giou_weight = giou_weight
+        
+    def giou_loss(self, preds, bbox, eps=1e-7, reduction='mean'):
+        '''
+        :param preds:[[x1,y1,x2,y2], [x1,y1,x2,y2],,,]
+        :param bbox:[[x1,y1,x2,y2], [x1,y1,x2,y2],,,]
+        :return: loss
+        '''
+        ix1 = fluid.layers.elementwise_max(preds[:, 0], bbox[:, 0])
+        iy1 = fluid.layers.elementwise_max(preds[:, 1], bbox[:, 1])
+        ix2 = fluid.layers.elementwise_min(preds[:, 2], bbox[:, 2])
+        iy2 = fluid.layers.elementwise_min(preds[:, 3], bbox[:, 3])
+
+        iw = fluid.layers.clip(ix2 - ix1 + 1e-3, 0., 1e10)
+        ih = fluid.layers.clip(iy2 - iy1 + 1e-3, 0., 1e10)
+
+        # overlap
+        inters = iw * ih
+
+        # union
+        uni = (preds[:, 2] - preds[:, 0] + 1e-3) * (preds[:, 3] - preds[:, 1] + 1e-3
+            ) + (bbox[:, 2] - bbox[:, 0] + 1e-3) * (
+            bbox[:, 3] - bbox[:, 1] + 1e-3) - inters + eps
+
+        # ious
+        ious = inters / uni
+
+        ex1 = fluid.layers.elementwise_min(preds[:, 0], bbox[:, 0])
+        ey1 = fluid.layers.elementwise_min(preds[:, 1], bbox[:, 1])
+        ex2 = fluid.layers.elementwise_max(preds[:, 2], bbox[:, 2])
+        ey2 = fluid.layers.elementwise_max(preds[:, 3], bbox[:, 3])
+        ew = fluid.layers.clip(ex2 - ex1 + 1e-3, 0., 1e10)
+        eh = fluid.layers.clip(ey2 - ey1 + 1e-3, 0., 1e10)
+
+        # enclose erea
+        enclose = ew * eh + eps
+        giou = ious - (enclose - uni) / enclose
+
+        loss = 1 - giou
+
+        if reduction == 'mean':
+            loss = paddle.mean(loss)
+        elif reduction == 'sum':
+            loss = paddle.sum(loss)
+        else:
+            raise NotImplementedError
+        return loss
+
+    def forward(self, predicts, batch):
+        structure_probs = predicts['structure_probs']
+        structure_targets = batch[1].astype("int64")
+        structure_targets = structure_targets[:, 1:]
+        if len(batch) == 6:
+            structure_mask = batch[5].astype("int64")
+            structure_mask = structure_mask[:, 1:]
+            structure_mask = paddle.reshape(structure_mask, [-1])
+        structure_probs = paddle.reshape(structure_probs, [-1, structure_probs.shape[-1]])
+        structure_targets = paddle.reshape(structure_targets, [-1])
+        structure_loss = self.loss_func(structure_probs, structure_targets)
+        
+        if len(batch) == 6:
+             structure_loss = structure_loss * structure_mask
+            
+#         structure_loss = paddle.sum(structure_loss) * self.structure_weight
+        structure_loss = paddle.mean(structure_loss) * self.structure_weight
+        
+        loc_preds = predicts['loc_preds']
+        loc_targets = batch[2].astype("float32")
+        loc_targets_mask = batch[4].astype("float32")
+        loc_targets = loc_targets[:, 1:, :]
+        loc_targets_mask = loc_targets_mask[:, 1:, :]
+        loc_loss = F.mse_loss(loc_preds * loc_targets_mask, loc_targets) * self.loc_weight
+        if self.use_giou:
+            loc_loss_giou = self.giou_loss(loc_preds * loc_targets_mask, loc_targets) * self.giou_weight
+            total_loss = structure_loss + loc_loss + loc_loss_giou
+            return {'loss':total_loss, "structure_loss":structure_loss, "loc_loss":loc_loss, "loc_loss_giou":loc_loss_giou}
+        else:
+            total_loss = structure_loss + loc_loss            
+            return {'loss':total_loss, "structure_loss":structure_loss, "loc_loss":loc_loss}
\ No newline at end of file

ppocr/metrics/__init__.py

@@ -19,20 +19,23 @@ from __future__ import unicode_literals
 
 import copy
 
-__all__ = ['build_metric']
+__all__ = ["build_metric"]
 
+from .det_metric import DetMetric
+from .rec_metric import RecMetric
+from .cls_metric import ClsMetric
+from .e2e_metric import E2EMetric
+from .distillation_metric import DistillationMetric
+from .table_metric import TableMetric
 
 def build_metric(config):
-    from .det_metric import DetMetric
-    from .rec_metric import RecMetric
-    from .cls_metric import ClsMetric
-    from .e2e_metric import E2EMetric
-
-    support_dict = ['DetMetric', 'RecMetric', 'ClsMetric', 'E2EMetric']
+    support_dict = [
+        "DetMetric", "RecMetric", "ClsMetric", "E2EMetric", "DistillationMetric", "TableMetric"
+    ]
 
     config = copy.deepcopy(config)
-    module_name = config.pop('name')
+    module_name = config.pop("name")
     assert module_name in support_dict, Exception(
-        'metric only support {}'.format(support_dict))
+        "metric only support {}".format(support_dict))
     module_class = eval(module_name)(**config)
     return module_class

ppocr/metrics/distillation_metric.py

+# 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.
+
+import importlib
+import copy
+
+from .rec_metric import RecMetric
+from .det_metric import DetMetric
+from .e2e_metric import E2EMetric
+from .cls_metric import ClsMetric
+
+
+class DistillationMetric(object):
+    def __init__(self,
+                 key=None,
+                 base_metric_name="RecMetric",
+                 main_indicator='acc',
+                 **kwargs):
+        self.main_indicator = main_indicator
+        self.key = key
+        self.main_indicator = main_indicator
+        self.base_metric_name = base_metric_name
+        self.kwargs = kwargs
+        self.metrics = None
+
+    def _init_metrcis(self, preds):
+        self.metrics = dict()
+        mod = importlib.import_module(__name__)
+        for key in preds:
+            self.metrics[key] = getattr(mod, self.base_metric_name)(
+                main_indicator=self.main_indicator, **self.kwargs)
+            self.metrics[key].reset()
+
+    def __call__(self, preds, *args, **kwargs):
+        assert isinstance(preds, dict)
+        if self.metrics is None:
+            self._init_metrcis(preds)
+        output = dict()
+        for key in preds:
+            metric = self.metrics[key].__call__(preds[key], *args, **kwargs)
+            for sub_key in metric:
+                output["{}_{}".format(key, sub_key)] = metric[sub_key]
+        return output
+
+    def get_metric(self):
+        """
+        return metrics {
+                 'acc': 0,
+                 'norm_edit_dis': 0,
+            }
+        """
+        output = dict()
+        for key in self.metrics:
+            metric = self.metrics[key].get_metric()
+            # main indicator
+            if key == self.key:
+                output.update(metric)
+            else:
+                for sub_key in metric:
+                    output["{}_{}".format(key, sub_key)] = metric[sub_key]
+        return output
+
+    def reset(self):
+        for key in self.metrics:
+            self.metrics[key].reset()

ppocr/metrics/table_metric.py

+# 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.
+import numpy as np
+class TableMetric(object):
+    def __init__(self, main_indicator='acc', **kwargs):
+        self.main_indicator = main_indicator
+        self.reset()
+
+    def __call__(self, pred, batch, *args, **kwargs):
+        structure_probs = pred['structure_probs'].numpy()
+        structure_labels = batch[1]
+        correct_num = 0
+        all_num = 0
+        structure_probs = np.argmax(structure_probs, axis=2)
+        structure_labels = structure_labels[:, 1:]
+        batch_size = structure_probs.shape[0]
+        for bno in range(batch_size):
+            all_num += 1
+            if (structure_probs[bno] == structure_labels[bno]).all():
+                correct_num += 1
+        self.correct_num += correct_num
+        self.all_num += all_num
+        return {
+            'acc': correct_num * 1.0 / all_num,
+        }
+
+    def get_metric(self):
+        """
+        return metrics {
+                 'acc': 0,
+            }
+        """
+        acc = 1.0 * self.correct_num / self.all_num
+        self.reset()
+        return {'acc': acc}
+
+    def reset(self):
+        self.correct_num = 0
+        self.all_num = 0

ppocr/modeling/architectures/__init__.py

@@ -13,12 +13,20 @@
 # limitations under the License.
 
 import copy
+import importlib
+
+from .base_model import BaseModel
+from .distillation_model import DistillationModel
 
 __all__ = ['build_model']
 
+
 def build_model(config):
-    from .base_model import BaseModel
-    
     config = copy.deepcopy(config)
-    module_class = BaseModel(config)
-    return module_class
\ No newline at end of file
+    if not "name" in config:
+        arch = BaseModel(config)
+    else:
+        name = config.pop("name")
+        mod = importlib.import_module(__name__)
+        arch = getattr(mod, name)(config)
+    return arch

ppocr/modeling/architectures/base_model.py

-# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -32,7 +32,6 @@ class BaseModel(nn.Layer):
             config (dict): the super parameters for module.
         """
         super(BaseModel, self).__init__()
-
         in_channels = config.get('in_channels', 3)
         model_type = config['model_type']
         # build transfrom,
@@ -68,14 +67,20 @@ class BaseModel(nn.Layer):
         config["Head"]['in_channels'] = in_channels
         self.head = build_head(config["Head"])
 
+        self.return_all_feats = config.get("return_all_feats", False)
+
     def forward(self, x, data=None):
+        y = dict()
         if self.use_transform:
             x = self.transform(x)
         x = self.backbone(x)
+        y["backbone_out"] = x
         if self.use_neck:
             x = self.neck(x)
-        if data is None:
-            x = self.head(x)
+        y["neck_out"] = x
+        x = self.head(x, targets=data)
+        y["head_out"] = x
+        if self.return_all_feats:
+            return y
         else:
-            x = self.head(x, data)
-        return x
+            return x

ppocr/modeling/architectures/distillation_model.py

+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# 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
+from __future__ import print_function
+
+from paddle import nn
+from ppocr.modeling.transforms import build_transform
+from ppocr.modeling.backbones import build_backbone
+from ppocr.modeling.necks import build_neck
+from ppocr.modeling.heads import build_head
+from .base_model import BaseModel
+from ppocr.utils.save_load import init_model
+
+__all__ = ['DistillationModel']
+
+
+class DistillationModel(nn.Layer):
+    def __init__(self, config):
+        """
+        the module for OCR distillation.
+        args:
+            config (dict): the super parameters for module.
+        """
+        super().__init__()
+        self.model_list = []
+        self.model_name_list = []
+        for key in config["Models"]:
+            model_config = config["Models"][key]
+            freeze_params = False
+            pretrained = None
+            if "freeze_params" in model_config:
+                freeze_params = model_config.pop("freeze_params")
+            if "pretrained" in model_config:
+                pretrained = model_config.pop("pretrained")
+            model = BaseModel(model_config)
+            if pretrained is not None:
+                init_model(model, path=pretrained)
+            if freeze_params:
+                for param in model.parameters():
+                    param.trainable = False
+            self.model_list.append(self.add_sublayer(key, model))
+            self.model_name_list.append(key)
+
+    def forward(self, x):
+        result_dict = dict()
+        for idx, model_name in enumerate(self.model_name_list):
+            result_dict[model_name] = self.model_list[idx](x)
+        return result_dict

ppocr/modeling/backbones/__init__.py

@@ -29,6 +29,10 @@ def build_backbone(config, model_type):
     elif model_type == 'e2e':
         from .e2e_resnet_vd_pg import ResNet
         support_dict = ['ResNet']
+    elif model_type == "table":
+        from .table_resnet_vd import ResNet
+        from .table_mobilenet_v3 import MobileNetV3
+        support_dict = ['ResNet', 'MobileNetV3']
     else:
         raise NotImplementedError
 

ppocr/modeling/backbones/det_mobilenet_v3.py

@@ -102,8 +102,7 @@ class MobileNetV3(nn.Layer):
             padding=1,
             groups=1,
             if_act=True,
-            act='hardswish',
-            name='conv1')
+            act='hardswish')
 
         self.stages = []
         self.out_channels = []
@@ -125,8 +124,7 @@ class MobileNetV3(nn.Layer):
                     kernel_size=k,
                     stride=s,
                     use_se=se,
-                    act=nl,
-                    name="conv" + str(i + 2)))
+                    act=nl))
             inplanes = make_divisible(scale * c)
             i += 1
         block_list.append(
@@ -138,8 +136,7 @@ class MobileNetV3(nn.Layer):
                 padding=0,
                 groups=1,
                 if_act=True,
-                act='hardswish',
-                name='conv_last'))
+                act='hardswish'))
         self.stages.append(nn.Sequential(*block_list))
         self.out_channels.append(make_divisible(scale * cls_ch_squeeze))
         for i, stage in enumerate(self.stages):
@@ -163,8 +160,7 @@ class ConvBNLayer(nn.Layer):
                  padding,
                  groups=1,
                  if_act=True,
-                 act=None,
-                 name=None):
+                 act=None):
         super(ConvBNLayer, self).__init__()
         self.if_act = if_act
         self.act = act
@@ -175,16 +171,9 @@ class ConvBNLayer(nn.Layer):
             stride=stride,
             padding=padding,
             groups=groups,
-            weight_attr=ParamAttr(name=name + '_weights'),
             bias_attr=False)
 
-        self.bn = nn.BatchNorm(
-            num_channels=out_channels,
-            act=None,
-            param_attr=ParamAttr(name=name + "_bn_scale"),
-            bias_attr=ParamAttr(name=name + "_bn_offset"),
-            moving_mean_name=name + "_bn_mean",
-            moving_variance_name=name + "_bn_variance")
+        self.bn = nn.BatchNorm(num_channels=out_channels, act=None)
 
     def forward(self, x):
         x = self.conv(x)
@@ -209,8 +198,7 @@ class ResidualUnit(nn.Layer):
                  kernel_size,
                  stride,
                  use_se,
-                 act=None,
-                 name=''):
+                 act=None):
         super(ResidualUnit, self).__init__()
         self.if_shortcut = stride == 1 and in_channels == out_channels
         self.if_se = use_se
@@ -222,8 +210,7 @@ class ResidualUnit(nn.Layer):
             stride=1,
             padding=0,
             if_act=True,
-            act=act,
-            name=name + "_expand")
+            act=act)
         self.bottleneck_conv = ConvBNLayer(
             in_channels=mid_channels,
             out_channels=mid_channels,
@@ -232,10 +219,9 @@ class ResidualUnit(nn.Layer):
             padding=int((kernel_size - 1) // 2),
             groups=mid_channels,
             if_act=True,
-            act=act,
-            name=name + "_depthwise")
+            act=act)
         if self.if_se:
-            self.mid_se = SEModule(mid_channels, name=name + "_se")
+            self.mid_se = SEModule(mid_channels)
         self.linear_conv = ConvBNLayer(
             in_channels=mid_channels,
             out_channels=out_channels,
@@ -243,8 +229,7 @@ class ResidualUnit(nn.Layer):
             stride=1,
             padding=0,
             if_act=False,
-            act=None,
-            name=name + "_linear")
+            act=None)
 
     def forward(self, inputs):
         x = self.expand_conv(inputs)
@@ -258,7 +243,7 @@ class ResidualUnit(nn.Layer):
 
 
 class SEModule(nn.Layer):
-    def __init__(self, in_channels, reduction=4, name=""):
+    def __init__(self, in_channels, reduction=4):
         super(SEModule, self).__init__()
         self.avg_pool = nn.AdaptiveAvgPool2D(1)
         self.conv1 = nn.Conv2D(
@@ -266,17 +251,13 @@ class SEModule(nn.Layer):
             out_channels=in_channels // reduction,
             kernel_size=1,
             stride=1,
-            padding=0,
-            weight_attr=ParamAttr(name=name + "_1_weights"),
-            bias_attr=ParamAttr(name=name + "_1_offset"))
+            padding=0)
         self.conv2 = nn.Conv2D(
             in_channels=in_channels // reduction,
             out_channels=in_channels,
             kernel_size=1,
             stride=1,
-            padding=0,
-            weight_attr=ParamAttr(name + "_2_weights"),
-            bias_attr=ParamAttr(name=name + "_2_offset"))
+            padding=0)
 
     def forward(self, inputs):
         outputs = self.avg_pool(inputs)

ppocr/modeling/backbones/rec_mobilenet_v3.py

@@ -96,8 +96,7 @@ class MobileNetV3(nn.Layer):
             padding=1,
             groups=1,
             if_act=True,
-            act='hardswish',
-            name='conv1')
+            act='hardswish')
         i = 0
         block_list = []
         inplanes = make_divisible(inplanes * scale)
@@ -110,8 +109,7 @@ class MobileNetV3(nn.Layer):
                     kernel_size=k,
                     stride=s,
                     use_se=se,
-                    act=nl,
-                    name='conv' + str(i + 2)))
+                    act=nl))
             inplanes = make_divisible(scale * c)
             i += 1
         self.blocks = nn.Sequential(*block_list)
@@ -124,8 +122,7 @@ class MobileNetV3(nn.Layer):
             padding=0,
             groups=1,
             if_act=True,
-            act='hardswish',
-            name='conv_last')
+            act='hardswish')
 
         self.pool = nn.MaxPool2D(kernel_size=2, stride=2, padding=0)
         self.out_channels = make_divisible(scale * cls_ch_squeeze)

ppocr/modeling/backbones/table_mobilenet_v3.py

+# 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.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+
+__all__ = ['MobileNetV3']
+
+
+def make_divisible(v, divisor=8, min_value=None):
+    if min_value is None:
+        min_value = divisor
+    new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
+    if new_v < 0.9 * v:
+        new_v += divisor
+    return new_v
+
+
+class MobileNetV3(nn.Layer):
+    def __init__(self,
+                 in_channels=3,
+                 model_name='large',
+                 scale=0.5,
+                 disable_se=False,
+                 **kwargs):
+        """
+        the MobilenetV3 backbone network for detection module.
+        Args:
+            params(dict): the super parameters for build network
+        """
+        super(MobileNetV3, self).__init__()
+
+        self.disable_se = disable_se
+
+        if model_name == "large":
+            cfg = [
+                # k, exp, c,  se,     nl,  s,
+                [3, 16, 16, False, 'relu', 1],
+                [3, 64, 24, False, 'relu', 2],
+                [3, 72, 24, False, 'relu', 1],
+                [5, 72, 40, True, 'relu', 2],
+                [5, 120, 40, True, 'relu', 1],
+                [5, 120, 40, True, 'relu', 1],
+                [3, 240, 80, False, 'hardswish', 2],
+                [3, 200, 80, False, 'hardswish', 1],
+                [3, 184, 80, False, 'hardswish', 1],
+                [3, 184, 80, False, 'hardswish', 1],
+                [3, 480, 112, True, 'hardswish', 1],
+                [3, 672, 112, True, 'hardswish', 1],
+                [5, 672, 160, True, 'hardswish', 2],
+                [5, 960, 160, True, 'hardswish', 1],
+                [5, 960, 160, True, 'hardswish', 1],
+            ]
+            cls_ch_squeeze = 960
+        elif model_name == "small":
+            cfg = [
+                # k, exp, c,  se,     nl,  s,
+                [3, 16, 16, True, 'relu', 2],
+                [3, 72, 24, False, 'relu', 2],
+                [3, 88, 24, False, 'relu', 1],
+                [5, 96, 40, True, 'hardswish', 2],
+                [5, 240, 40, True, 'hardswish', 1],
+                [5, 240, 40, True, 'hardswish', 1],
+                [5, 120, 48, True, 'hardswish', 1],
+                [5, 144, 48, True, 'hardswish', 1],
+                [5, 288, 96, True, 'hardswish', 2],
+                [5, 576, 96, True, 'hardswish', 1],
+                [5, 576, 96, True, 'hardswish', 1],
+            ]
+            cls_ch_squeeze = 576
+        else:
+            raise NotImplementedError("mode[" + model_name +
+                                      "_model] is not implemented!")
+
+        supported_scale = [0.35, 0.5, 0.75, 1.0, 1.25]
+        assert scale in supported_scale, \
+            "supported scale are {} but input scale is {}".format(supported_scale, scale)
+        inplanes = 16
+        # conv1
+        self.conv = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=make_divisible(inplanes * scale),
+            kernel_size=3,
+            stride=2,
+            padding=1,
+            groups=1,
+            if_act=True,
+            act='hardswish',
+            name='conv1')
+
+        self.stages = []
+        self.out_channels = []
+        block_list = []
+        i = 0
+        inplanes = make_divisible(inplanes * scale)
+        for (k, exp, c, se, nl, s) in cfg:
+            se = se and not self.disable_se
+            start_idx = 2 if model_name == 'large' else 0
+            if s == 2 and i > start_idx:
+                self.out_channels.append(inplanes)
+                self.stages.append(nn.Sequential(*block_list))
+                block_list = []
+            block_list.append(
+                ResidualUnit(
+                    in_channels=inplanes,
+                    mid_channels=make_divisible(scale * exp),
+                    out_channels=make_divisible(scale * c),
+                    kernel_size=k,
+                    stride=s,
+                    use_se=se,
+                    act=nl,
+                    name="conv" + str(i + 2)))
+            inplanes = make_divisible(scale * c)
+            i += 1
+        block_list.append(
+            ConvBNLayer(
+                in_channels=inplanes,
+                out_channels=make_divisible(scale * cls_ch_squeeze),
+                kernel_size=1,
+                stride=1,
+                padding=0,
+                groups=1,
+                if_act=True,
+                act='hardswish',
+                name='conv_last'))
+        self.stages.append(nn.Sequential(*block_list))
+        self.out_channels.append(make_divisible(scale * cls_ch_squeeze))
+        for i, stage in enumerate(self.stages):
+            self.add_sublayer(sublayer=stage, name="stage{}".format(i))
+
+    def forward(self, x):
+        x = self.conv(x)
+        out_list = []
+        for stage in self.stages:
+            x = stage(x)
+            out_list.append(x)
+        return out_list
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(self,
+                 in_channels,
+                 out_channels,
+                 kernel_size,
+                 stride,
+                 padding,
+                 groups=1,
+                 if_act=True,
+                 act=None,
+                 name=None):
+        super(ConvBNLayer, self).__init__()
+        self.if_act = if_act
+        self.act = act
+        self.conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=padding,
+            groups=groups,
+            weight_attr=ParamAttr(name=name + '_weights'),
+            bias_attr=False)
+
+        self.bn = nn.BatchNorm(
+            num_channels=out_channels,
+            act=None,
+            param_attr=ParamAttr(name=name + "_bn_scale"),
+            bias_attr=ParamAttr(name=name + "_bn_offset"),
+            moving_mean_name=name + "_bn_mean",
+            moving_variance_name=name + "_bn_variance")
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.bn(x)
+        if self.if_act:
+            if self.act == "relu":
+                x = F.relu(x)
+            elif self.act == "hardswish":
+                x = F.hardswish(x)
+            else:
+                print("The activation function({}) is selected incorrectly.".
+                      format(self.act))
+                exit()
+        return x
+
+
+class ResidualUnit(nn.Layer):
+    def __init__(self,
+                 in_channels,
+                 mid_channels,
+                 out_channels,
+                 kernel_size,
+                 stride,
+                 use_se,
+                 act=None,
+                 name=''):
+        super(ResidualUnit, self).__init__()
+        self.if_shortcut = stride == 1 and in_channels == out_channels
+        self.if_se = use_se
+
+        self.expand_conv = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=mid_channels,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            if_act=True,
+            act=act,
+            name=name + "_expand")
+        self.bottleneck_conv = ConvBNLayer(
+            in_channels=mid_channels,
+            out_channels=mid_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=int((kernel_size - 1) // 2),
+            groups=mid_channels,
+            if_act=True,
+            act=act,
+            name=name + "_depthwise")
+        if self.if_se:
+            self.mid_se = SEModule(mid_channels, name=name + "_se")
+        self.linear_conv = ConvBNLayer(
+            in_channels=mid_channels,
+            out_channels=out_channels,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            if_act=False,
+            act=None,
+            name=name + "_linear")
+
+    def forward(self, inputs):
+        x = self.expand_conv(inputs)
+        x = self.bottleneck_conv(x)
+        if self.if_se:
+            x = self.mid_se(x)
+        x = self.linear_conv(x)
+        if self.if_shortcut:
+            x = paddle.add(inputs, x)
+        return x
+
+
+class SEModule(nn.Layer):
+    def __init__(self, in_channels, reduction=4, name=""):
+        super(SEModule, self).__init__()
+        self.avg_pool = nn.AdaptiveAvgPool2D(1)
+        self.conv1 = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=in_channels // reduction,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            weight_attr=ParamAttr(name=name + "_1_weights"),
+            bias_attr=ParamAttr(name=name + "_1_offset"))
+        self.conv2 = nn.Conv2D(
+            in_channels=in_channels // reduction,
+            out_channels=in_channels,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            weight_attr=ParamAttr(name + "_2_weights"),
+            bias_attr=ParamAttr(name=name + "_2_offset"))
+
+    def forward(self, inputs):
+        outputs = self.avg_pool(inputs)
+        outputs = self.conv1(outputs)
+        outputs = F.relu(outputs)
+        outputs = self.conv2(outputs)
+        outputs = F.hardsigmoid(outputs, slope=0.2, offset=0.5)
+        return inputs * outputs
\ No newline at end of file

ppocr/modeling/backbones/table_resnet_vd.py

+# 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.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import ParamAttr
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+__all__ = ["ResNet"]
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+            self,
+            in_channels,
+            out_channels,
+            kernel_size,
+            stride=1,
+            groups=1,
+            is_vd_mode=False,
+            act=None,
+            name=None, ):
+        super(ConvBNLayer, self).__init__()
+
+        self.is_vd_mode = is_vd_mode
+        self._pool2d_avg = nn.AvgPool2D(
+            kernel_size=2, stride=2, padding=0, ceil_mode=True)
+        self._conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=(kernel_size - 1) // 2,
+            groups=groups,
+            weight_attr=ParamAttr(name=name + "_weights"),
+            bias_attr=False)
+        if name == "conv1":
+            bn_name = "bn_" + name
+        else:
+            bn_name = "bn" + name[3:]
+        self._batch_norm = nn.BatchNorm(
+            out_channels,
+            act=act,
+            param_attr=ParamAttr(name=bn_name + '_scale'),
+            bias_attr=ParamAttr(bn_name + '_offset'),
+            moving_mean_name=bn_name + '_mean',
+            moving_variance_name=bn_name + '_variance')
+
+    def forward(self, inputs):
+        if self.is_vd_mode:
+            inputs = self._pool2d_avg(inputs)
+        y = self._conv(inputs)
+        y = self._batch_norm(y)
+        return y
+
+
+class BottleneckBlock(nn.Layer):
+    def __init__(self,
+                 in_channels,
+                 out_channels,
+                 stride,
+                 shortcut=True,
+                 if_first=False,
+                 name=None):
+        super(BottleneckBlock, self).__init__()
+
+        self.conv0 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=1,
+            act='relu',
+            name=name + "_branch2a")
+        self.conv1 = ConvBNLayer(
+            in_channels=out_channels,
+            out_channels=out_channels,
+            kernel_size=3,
+            stride=stride,
+            act='relu',
+            name=name + "_branch2b")
+        self.conv2 = ConvBNLayer(
+            in_channels=out_channels,
+            out_channels=out_channels * 4,
+            kernel_size=1,
+            act=None,
+            name=name + "_branch2c")
+
+        if not shortcut:
+            self.short = ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=out_channels * 4,
+                kernel_size=1,
+                stride=1,
+                is_vd_mode=False if if_first else True,
+                name=name + "_branch1")
+
+        self.shortcut = shortcut
+
+    def forward(self, inputs):
+        y = self.conv0(inputs)
+        conv1 = self.conv1(y)
+        conv2 = self.conv2(conv1)
+
+        if self.shortcut:
+            short = inputs
+        else:
+            short = self.short(inputs)
+        y = paddle.add(x=short, y=conv2)
+        y = F.relu(y)
+        return y
+
+
+class BasicBlock(nn.Layer):
+    def __init__(self,
+                 in_channels,
+                 out_channels,
+                 stride,
+                 shortcut=True,
+                 if_first=False,
+                 name=None):
+        super(BasicBlock, self).__init__()
+        self.stride = stride
+        self.conv0 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=3,
+            stride=stride,
+            act='relu',
+            name=name + "_branch2a")
+        self.conv1 = ConvBNLayer(
+            in_channels=out_channels,
+            out_channels=out_channels,
+            kernel_size=3,
+            act=None,
+            name=name + "_branch2b")
+
+        if not shortcut:
+            self.short = ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=out_channels,
+                kernel_size=1,
+                stride=1,
+                is_vd_mode=False if if_first else True,
+                name=name + "_branch1")
+
+        self.shortcut = shortcut
+
+    def forward(self, inputs):
+        y = self.conv0(inputs)
+        conv1 = self.conv1(y)
+
+        if self.shortcut:
+            short = inputs
+        else:
+            short = self.short(inputs)
+        y = paddle.add(x=short, y=conv1)
+        y = F.relu(y)
+        return y
+
+
+class ResNet(nn.Layer):
+    def __init__(self, in_channels=3, layers=50, **kwargs):
+        super(ResNet, self).__init__()
+        
+        self.layers = layers
+        supported_layers = [18, 34, 50, 101, 152, 200]
+        assert layers in supported_layers, \
+            "supported layers are {} but input layer is {}".format(
+                supported_layers, layers)
+
+        if layers == 18:
+            depth = [2, 2, 2, 2]
+        elif layers == 34 or layers == 50:
+            depth = [3, 4, 6, 3]
+        elif layers == 101:
+            depth = [3, 4, 23, 3]
+        elif layers == 152:
+            depth = [3, 8, 36, 3]
+        elif layers == 200:
+            depth = [3, 12, 48, 3]
+        num_channels = [64, 256, 512,
+                        1024] if layers >= 50 else [64, 64, 128, 256]
+        num_filters = [64, 128, 256, 512]
+
+        self.conv1_1 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=32,
+            kernel_size=3,
+            stride=2,
+            act='relu',
+            name="conv1_1")
+        self.conv1_2 = ConvBNLayer(
+            in_channels=32,
+            out_channels=32,
+            kernel_size=3,
+            stride=1,
+            act='relu',
+            name="conv1_2")
+        self.conv1_3 = ConvBNLayer(
+            in_channels=32,
+            out_channels=64,
+            kernel_size=3,
+            stride=1,
+            act='relu',
+            name="conv1_3")
+        self.pool2d_max = nn.MaxPool2D(kernel_size=3, stride=2, padding=1)
+
+        self.stages = []
+        self.out_channels = []
+        if layers >= 50:
+            for block in range(len(depth)):
+                block_list = []
+                shortcut = False
+                for i in range(depth[block]):
+                    if layers in [101, 152] and block == 2:
+                        if i == 0:
+                            conv_name = "res" + str(block + 2) + "a"
+                        else:
+                            conv_name = "res" + str(block + 2) + "b" + str(i)
+                    else:
+                        conv_name = "res" + str(block + 2) + chr(97 + i)
+                    bottleneck_block = self.add_sublayer(
+                        'bb_%d_%d' % (block, i),
+                        BottleneckBlock(
+                            in_channels=num_channels[block]
+                            if i == 0 else num_filters[block] * 4,
+                            out_channels=num_filters[block],
+                            stride=2 if i == 0 and block != 0 else 1,
+                            shortcut=shortcut,
+                            if_first=block == i == 0,
+                            name=conv_name))
+                    shortcut = True
+                    block_list.append(bottleneck_block)
+                self.out_channels.append(num_filters[block] * 4)
+                self.stages.append(nn.Sequential(*block_list))
+        else:
+            for block in range(len(depth)):
+                block_list = []
+                shortcut = False
+                for i in range(depth[block]):
+                    conv_name = "res" + str(block + 2) + chr(97 + i)
+                    basic_block = self.add_sublayer(
+                        'bb_%d_%d' % (block, i),
+                        BasicBlock(
+                            in_channels=num_channels[block]
+                            if i == 0 else num_filters[block],
+                            out_channels=num_filters[block],
+                            stride=2 if i == 0 and block != 0 else 1,
+                            shortcut=shortcut,
+                            if_first=block == i == 0,
+                            name=conv_name))
+                    shortcut = True
+                    block_list.append(basic_block)
+                self.out_channels.append(num_filters[block])
+                self.stages.append(nn.Sequential(*block_list))
+
+    def forward(self, inputs):
+        y = self.conv1_1(inputs)
+        y = self.conv1_2(y)
+        y = self.conv1_3(y)
+        y = self.pool2d_max(y)
+        out = []
+        for block in self.stages:
+            y = block(y)
+            out.append(y)
+        return out

ppocr/modeling/heads/__init__.py

@@ -31,8 +31,10 @@ def build_head(config):
     from .cls_head import ClsHead
     support_dict = [
         'DBHead', 'EASTHead', 'SASTHead', 'CTCHead', 'ClsHead', 'AttentionHead',
-        'SRNHead', 'PGHead']
+        'SRNHead', 'PGHead', 'TableAttentionHead']
 
+    #table head
+    from .table_att_head import TableAttentionHead
 
     module_name = config.pop('name')
     assert module_name in support_dict, Exception('head only support {}'.format(

ppocr/modeling/heads/cls_head.py

@@ -43,7 +43,7 @@ class ClsHead(nn.Layer):
                 initializer=nn.initializer.Uniform(-stdv, stdv)),
             bias_attr=ParamAttr(name="fc_0.b_0"), )
 
-    def forward(self, x):
+    def forward(self, x, targets=None):
         x = self.pool(x)
         x = paddle.reshape(x, shape=[x.shape[0], x.shape[1]])
         x = self.fc(x)