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

ppocr/data/__init__.py

@@ -34,6 +34,7 @@ import paddle.distributed as dist
 from ppocr.data.imaug import transform, create_operators
 from ppocr.data.simple_dataset import SimpleDataSet
 from ppocr.data.lmdb_dataset import LMDBDataSet
+from ppocr.data.pgnet_dataset import PGDataSet
 
 __all__ = ['build_dataloader', 'transform', 'create_operators']
 
@@ -54,7 +55,7 @@ signal.signal(signal.SIGTERM, term_mp)
 def build_dataloader(config, mode, device, logger, seed=None):
     config = copy.deepcopy(config)
 
-    support_dict = ['SimpleDataSet', 'LMDBDataSet']
+    support_dict = ['SimpleDataSet', 'LMDBDataSet', 'PGDataSet']
     module_name = config[mode]['dataset']['name']
     assert module_name in support_dict, Exception(
         'DataSet only support {}'.format(support_dict))
@@ -72,14 +73,14 @@ def build_dataloader(config, mode, device, logger, seed=None):
     else:
         use_shared_memory = True
     if mode == "Train":
-        #Distribute data to multiple cards
+        # Distribute data to multiple cards
         batch_sampler = DistributedBatchSampler(
             dataset=dataset,
             batch_size=batch_size,
             shuffle=shuffle,
             drop_last=drop_last)
     else:
-        #Distribute data to single card
+        # Distribute data to single card
         batch_sampler = BatchSampler(
             dataset=dataset,
             batch_size=batch_size,

ppocr/data/imaug/__init__.py

@@ -28,6 +28,7 @@ from .label_ops import *
 
 from .east_process import *
 from .sast_process import *
+from .pg_process import *
 
 
 def transform(data, ops=None):

ppocr/data/imaug/randaugment.py

@@ -117,13 +117,16 @@ class RawRandAugment(object):
 class RandAugment(RawRandAugment):
     """ RandAugment wrapper to auto fit different img types """
 
-    def __init__(self, *args, **kwargs):
+    def __init__(self, prob=0.5, *args, **kwargs):
+        self.prob = prob
         if six.PY2:
             super(RandAugment, self).__init__(*args, **kwargs)
         else:
             super().__init__(*args, **kwargs)
 
     def __call__(self, data):
+        if np.random.rand() > self.prob:
+            return data
         img = data['image']
         if not isinstance(img, Image.Image):
             img = np.ascontiguousarray(img)

ppocr/losses/__init__.py

@@ -29,10 +29,11 @@ def build_loss(config):
     # cls loss
     from .cls_loss import ClsLoss
 
+    # e2e loss
+    from .e2e_pg_loss import PGLoss
     support_dict = [
         'DBLoss', 'EASTLoss', 'SASTLoss', 'CTCLoss', 'ClsLoss', 'AttentionLoss',
-        'SRNLoss'
-    ]
+        'SRNLoss', 'PGLoss']
 
     config = copy.deepcopy(config)
     module_name = config.pop('name')

ppocr/losses/det_basic_loss.py

@@ -200,6 +200,6 @@ def ohem_batch(scores, gt_texts, training_masks, ohem_ratio):
                 i, :, :], ohem_ratio))
 
     selected_masks = np.concatenate(selected_masks, 0)
-    selected_masks = paddle.to_variable(selected_masks)
+    selected_masks = paddle.to_tensor(selected_masks)
 
     return selected_masks