add aster rosetta

configs/rec/rec_mv3_none_bilstm_ctc.yml

@@ -27,7 +27,7 @@ Optimizer:
   weight_decay: 0.0
 
 LRScheduler:
-  name: CosineAnnealingLR
+  name: ConstLR
   warmup_epoch: 0
 
 Architecture:

configs/rec/rec_mv3_none_none_ctc.yml

@@ -1,34 +1,34 @@
 Global:
-  use_gpu: True
+  device: gpu
   epoch_num: 72
   log_smooth_window: 20
   print_batch_step: 10
-  output_dir: ./output/rec/mv3_none_none_ctc/
-  save_epoch_step: 3
-  # evaluation is run every 2000 iterations
-  eval_batch_step: [0, 2000]
-  cal_metric_during_train: True
+  output_dir: ./output/rec/mv3_none_none_ctc
+  eval_epoch_step: [0, 1]
+  cal_metric_during_train: true
   pretrained_model:
   checkpoints:
-  save_inference_dir:
-  use_visualdl: False
-  infer_img: doc/imgs_words_en/word_10.png
-  # for data or label process
-  character_dict_path:
-  max_text_length: 25
-  infer_mode: False
-  use_space_char: False
-  save_res_path: ./output/rec/predicts_mv3_none_none_ctc.txt
+  use_tensorboard: false
+  infer_mode: false
+  infer_img: doc/imgs_words/en/word_1.png
+  character_dict_path: &character_dict_path
+  max_text_length: &max_text_length 25
+  use_space_char: &use_space_char False
+
+Export:
+  export_dir:
+  export_shape: [ 1, 3, 32, 100 ]
+  dynamic_axes: [ 0, 2, 3 ]
 
 Optimizer:
   name: Adam
-  beta1: 0.9
-  beta2: 0.999
-  lr:
-    learning_rate: 0.0005
-  regularizer:
-    name: 'L2'
-    factor: 0
+  lr: 0.0005
+  weight_decay: 0
+
+LRScheduler:
+  name: ConstLR
+  warmup_epoch: 0
+
 
 Architecture:
   model_type: rec
@@ -50,6 +50,8 @@ Loss:
 
 PostProcess:
   name: CTCLabelDecode
+  character_dict_path: *character_dict_path
+  use_space_char: *use_space_char
 
 Metric:
   name: RecMetric

configs/rec/rec_mv3_tps_bilstm_att.yml

@@ -1,35 +1,34 @@
 Global:
-  use_gpu: True
+  device: gpu
   epoch_num: 72
   log_smooth_window: 20
   print_batch_step: 10
-  output_dir: ./output/rec/rec_mv3_tps_bilstm_att/
-  save_epoch_step: 3
-  # evaluation is run every 5000 iterations after the 4000th iteration
-  eval_batch_step: [0, 2000]
-  cal_metric_during_train: True
+  output_dir: ./output/rec/rec_mv3_tps_bilstm_att
+  eval_epoch_step: [0, 1]
+  cal_metric_during_train: true
   pretrained_model:
   checkpoints:
-  save_inference_dir:
-  use_visualdl: False
-  infer_img: doc/imgs_words/ch/word_1.jpg
-  # for data or label process
-  character_dict_path:
-  max_text_length: 25
-  infer_mode: False
-  use_space_char: False
-  save_res_path: ./output/rec/predicts_mv3_tps_bilstm_att.txt
+  use_tensorboard: false
+  infer_mode: false
+  infer_img: doc/imgs_words/en/word_1.png
+  character_dict_path: &character_dict_path
+  max_text_length: &max_text_length 25
+  use_space_char: &use_space_char False
 
 
+Export:
+  export_dir:
+  export_shape: [ 1, 3, 32, 100 ]
+  dynamic_axes: [ 0, 2, 3 ]
+
 Optimizer:
   name: Adam
-  beta1: 0.9
-  beta2: 0.999
-  lr:
-    learning_rate: 0.0005
-  regularizer:
-    name: 'L2'
-    factor: 0.00001
+  lr: 0.0005
+  weight_decay: 0.00001
+
+LRScheduler:
+  name: ConstLR
+  warmup_epoch: 0
 
 Architecture:
   model_type: rec
@@ -57,6 +56,8 @@ Loss:
 
 PostProcess:
   name: AttnLabelDecode
+  character_dict_path: *character_dict_path
+  use_space_char: *use_space_char
 
 Metric:
   name: RecMetric

configs/rec/rec_mv3_tps_bilstm_ctc.yml

@@ -3,7 +3,7 @@ Global:
   epoch_num: 72
   log_smooth_window: 20
   print_batch_step: 10
-  output_dir: ./output/rec/mv3_tps_bilstm_ctc
+  output_dir: ./output/rec/rec_mv3_tps_bilstm_ctc
   eval_epoch_step: [0, 1]
   cal_metric_during_train: true
   pretrained_model:
@@ -26,7 +26,7 @@ Optimizer:
   weight_decay: 0.0
 
 LRScheduler:
-  name: CosineAnnealingLR
+  name: ConstLR
   warmup_epoch: 0
 
 Architecture:

ppocr/modeling/heads/rec_att_head.py

@@ -37,14 +37,15 @@ def _char_to_onehot(self, input_char, onehot_dim):
         input_ont_hot = F.one_hot(input_char, onehot_dim)
         return input_ont_hot
 
-    def forward(self, inputs, targets=None, batch_max_length=25):
+    def forward(self, inputs, data=None, batch_max_length=25):
         batch_size = paddle.shape(inputs)[0]
         num_steps = batch_max_length
 
         hidden = paddle.zeros((batch_size, self.hidden_size))
         output_hiddens = []
 
-        if targets is not None:
+        if data is not None:
+            targets = data[1]
             for i in range(num_steps):
                 char_onehots = self._char_to_onehot(
                     targets[:, i], onehot_dim=self.num_classes)
@@ -76,7 +77,7 @@ def forward(self, inputs, targets=None, batch_max_length=25):
                 targets = next_input
         if not self.training:
             probs = paddle.nn.functional.softmax(probs, axis=2)
-        return probs
+        return {'res':probs}
 
 
 class AttentionGRUCell(nn.Layer):

readme.md

@@ -54,8 +54,10 @@
 
 | 模型 | 是否对齐 | 对齐误差| 配置文件 |
 |---|---|---|---|
+| rec_mv3_none_none_ctc   | X | 2.114354e-09 | [config](configs/rec/rec_mv3_none_none_ctc.yml) |
 | rec_mv3_none_bilstm_ctc   | X | 1.1861777e-09 | [config](configs/rec/rec_mv3_none_bilstm_ctc.yml) |
 | rec_mv3_tps_bilstm_ctc   | X | 1.1886948e-09 | [config](configs/rec/rec_mv3_tps_bilstm_ctc.yml) |
+| rec_mv3_tps_bilstm_att   | X | 1.8528418e-09 | [config](configs/rec/rec_mv3_tps_bilstm_att.yml) |
 
 ## TODO
 

torchocr/losses/rec_att_loss.py

@@ -8,10 +8,8 @@ def __init__(self, **kwargs):
         self.loss_func = nn.CrossEntropyLoss(weight=None, reduction='none')
 
     def forward(self, predicts, batch):
-        targets = batch[1].astype("int64")
-        label_lengths = batch[2].astype('int64')
-        batch_size, num_steps, num_classes = predicts.shape[0], predicts.shape[
-            1], predicts.shape[2]
+        predicts = predicts['res'][:, :-1]
+        targets = batch[1].long()[:,1:]
         assert len(targets.shape) == len(list(predicts.shape)) - 1, \
             "The target's shape and inputs's shape is [N, d] and [N, num_steps]"
 

torchocr/modeling/heads/__init__.py

@@ -8,11 +8,12 @@ def build_head(config):
     from .rec_ctc_head import CTCHead
     from .rec_multi_head import MultiHead
     from .rec_sar_head import SARHead
+    from .rec_att_head import AttentionHead
     # cls head
     from .cls_head import ClsHead
 
     support_dict = [
-        'MultiHead', 'SARHead', 'DBHead', 'CTCHead', 'ClsHead', 'PFHeadLocal'
+        'MultiHead', 'SARHead', 'DBHead', 'CTCHead', 'ClsHead', 'PFHeadLocal', 'AttentionHead'
     ]
 
     module_name = config.pop('name')