mindspore-lab · SamitHuang · Jul 3, 2023 · Jun 29, 2023 · Jul 3, 2023
diff --git a/configs/det/dbnet/db++_r50_icdar15.yaml b/configs/det/dbnet/db++_r50_icdar15.yaml
@@ -40,7 +40,7 @@ metric:
   main_indicator: f-score
 
 loss:
-  name: L1BalancedCELoss
+  name: DBLoss
   eps: 1.0e-6
   l1_scale: 10
   bce_scale: 5

diff --git a/configs/det/dbnet/db_mobilenetv3_icdar15.yaml b/configs/det/dbnet/db_mobilenetv3_icdar15.yaml
@@ -43,7 +43,7 @@ metric:
   main_indicator: f-score
 
 loss:
-  name: L1BalancedCELoss
+  name: DBLoss
   eps: 1.0e-6
   l1_scale: 10
   bce_scale: 5

diff --git a/configs/det/dbnet/db_r18_ctw1500.yaml b/configs/det/dbnet/db_r18_ctw1500.yaml
@@ -37,7 +37,7 @@ metric:
   main_indicator: f-score
 
 loss:
-  name: L1BalancedCELoss
+  name: DBLoss
   eps: 1.0e-6
   l1_scale: 10
   bce_scale: 5

diff --git a/configs/det/dbnet/db_r18_icdar15.yaml b/configs/det/dbnet/db_r18_icdar15.yaml
@@ -37,7 +37,7 @@ metric:
   main_indicator: f-score
 
 loss:
-  name: L1BalancedCELoss
+  name: DBLoss
   eps: 1.0e-6
   l1_scale: 10
   bce_scale: 5

diff --git a/configs/det/dbnet/db_r18_mlt2017.yaml b/configs/det/dbnet/db_r18_mlt2017.yaml
@@ -37,7 +37,7 @@ metric:
   main_indicator: f-score
 
 loss:
-  name: L1BalancedCELoss
+  name: DBLoss
   eps: 1.0e-6
   l1_scale: 10
   bce_scale: 5

diff --git a/configs/det/dbnet/db_r18_synthtext.yaml b/configs/det/dbnet/db_r18_synthtext.yaml
@@ -24,7 +24,7 @@ model:
     adaptive: True
 
 loss:
-  name: L1BalancedCELoss
+  name: DBLoss
   eps: 1.0e-6
   l1_scale: 10
   bce_scale: 5

diff --git a/configs/det/dbnet/db_r18_td500.yaml b/configs/det/dbnet/db_r18_td500.yaml
@@ -38,7 +38,7 @@ metric:
   main_indicator: f-score
 
 loss:
-  name: L1BalancedCELoss
+  name: DBLoss
   eps: 1.0e-6
   l1_scale: 10
   bce_scale: 5

diff --git a/configs/det/dbnet/db_r18_totaltext.yaml b/configs/det/dbnet/db_r18_totaltext.yaml
@@ -37,7 +37,7 @@ metric:
   main_indicator: f-score
 
 loss:
-  name: L1BalancedCELoss
+  name: DBLoss
   eps: 1.0e-6
   l1_scale: 10
   bce_scale: 5

diff --git a/configs/det/dbnet/db_r50_ctw1500.yaml b/configs/det/dbnet/db_r50_ctw1500.yaml
@@ -37,7 +37,7 @@ metric:
   main_indicator: f-score
 
 loss:
-  name: L1BalancedCELoss
+  name: DBLoss
   eps: 1.0e-6
   l1_scale: 10
   bce_scale: 5

diff --git a/configs/det/dbnet/db_r50_icdar15.yaml b/configs/det/dbnet/db_r50_icdar15.yaml
@@ -36,7 +36,7 @@ metric:
   main_indicator: f-score
 
 loss:
-  name: L1BalancedCELoss
+  name: DBLoss
   eps: 1.0e-6
   l1_scale: 10
   bce_scale: 5

diff --git a/configs/det/dbnet/db_r50_mlt2017.yaml b/configs/det/dbnet/db_r50_mlt2017.yaml
@@ -37,7 +37,7 @@ metric:
   main_indicator: f-score
 
 loss:
-  name: L1BalancedCELoss
+  name: DBLoss
   eps: 1.0e-6
   l1_scale: 10
   bce_scale: 5

diff --git a/configs/det/dbnet/db_r50_synthtext.yaml b/configs/det/dbnet/db_r50_synthtext.yaml
@@ -24,7 +24,7 @@ model:
     adaptive: True
 
 loss:
-  name: L1BalancedCELoss
+  name: DBLoss
   eps: 1.0e-6
   l1_scale: 10
   bce_scale: 5

diff --git a/configs/det/dbnet/db_r50_td500.yaml b/configs/det/dbnet/db_r50_td500.yaml
@@ -38,7 +38,7 @@ metric:
   main_indicator: f-score
 
 loss:
-  name: L1BalancedCELoss
+  name: DBLoss
   eps: 1.0e-6
   l1_scale: 10
   bce_scale: 5

diff --git a/configs/det/dbnet/db_r50_totaltext.yaml b/configs/det/dbnet/db_r50_totaltext.yaml
@@ -37,7 +37,7 @@ metric:
   main_indicator: f-score
 
 loss:
-  name: L1BalancedCELoss
+  name: DBLoss
   eps: 1.0e-6
   l1_scale: 10
   bce_scale: 5

diff --git a/docs/cn/tutorials/training_detection_custom_dataset.md b/docs/cn/tutorials/training_detection_custom_dataset.md
@@ -215,7 +215,7 @@ metric:
   main_indicator: f-score
 
 loss:
-  name: L1BalancedCELoss
+  name: DBLoss
   eps: 1.0e-6
   l1_scale: 10
   bce_scale: 5
@@ -234,7 +234,7 @@ optimizer:
   momentum: 0.9
   weight_decay: 1.0e-4
 ```
-它使用 `SGD` 优化器（在 `mindocr/optim/optim.factory.py` 中）和 `polynomial_decay`（在 `mindocr/scheduler/scheduler_factory.py` 中）作为学习率调整策略。损失函数是 `L1BalancedCELoss`（在 `mindocr/losses/det_loss.py` 中），评估指标是 `DetMetric`（在 `mindocr/metrics/det_metrics.py` 中）。
+它使用 `SGD` 优化器（在 `mindocr/optim/optim.factory.py` 中）和 `polynomial_decay`（在 `mindocr/scheduler/scheduler_factory.py` 中）作为学习率调整策略。损失函数是 `DBLoss`（在 `mindocr/losses/det_loss.py` 中），评估指标是 `DetMetric`（在 `mindocr/metrics/det_metrics.py` 中）。
 
 ## 3. 模型训练, 测试和推理
 

diff --git a/docs/cn/tutorials/yaml_configuration.md b/docs/cn/tutorials/yaml_configuration.md
@@ -99,7 +99,7 @@
 
 | 字段 | 用途 | 默认值 | 备注 |
 | :---- | :---- | :---- | :---- |
-| name | 损失函数类名 | - | 目前支持 L1BalancedCELoss, CTCLoss, AttentionLoss, PSEDiceLoss, EASTLoss and CrossEntropySmooth |
+| name | 损失函数类名 | - | 目前支持 DBLoss, CTCLoss, AttentionLoss, PSEDiceLoss, EASTLoss and CrossEntropySmooth |
 | pred_seq_len | 预测文本的长度 | 26 | 由网络架构决定 |
 | max_label_len | 最长标签长度 | 25 | 数值应小于网络预测文本的长度 |
 | batch_size | 单卡批量大小 | 32 | \ |

diff --git a/docs/en/tutorials/training_detection_custom_dataset.md b/docs/en/tutorials/training_detection_custom_dataset.md
@@ -214,7 +214,7 @@ metric:
   main_indicator: f-score
 
 loss:
-  name: L1BalancedCELoss
+  name: DBLoss
   eps: 1.0e-6
   l1_scale: 10
   bce_scale: 5
@@ -233,7 +233,7 @@ optimizer:
   momentum: 0.9
   weight_decay: 1.0e-4
 ```
-It uses `SGD` optimizer (in `mindocr/optim/optim.factory.py`) and `polynomial_decay` (in `mindocr/scheduler/scheduler_factory.py`) as the learning scheduler. The loss function is `L1BalancedCELoss` (in `mindocr/losses/det_loss.py`) and the evaluation metric is `DetMetric` ( in `mindocr/metrics/det_metrics.py`).
+It uses `SGD` optimizer (in `mindocr/optim/optim.factory.py`) and `polynomial_decay` (in `mindocr/scheduler/scheduler_factory.py`) as the learning scheduler. The loss function is `DBLoss` (in `mindocr/losses/det_loss.py`) and the evaluation metric is `DetMetric` ( in `mindocr/metrics/det_metrics.py`).
 
 
 ## 3. Model Training, Evaluation, and Inference

diff --git a/docs/en/tutorials/yaml_configuration.md b/docs/en/tutorials/yaml_configuration.md
@@ -97,7 +97,7 @@ Please see the code in [mindocr/losses](../../../mindocr/losses)
 
 | Parameter | Description | Default | Remarks |
 | :---- | :---- | :---- | :---- |
-| name | loss function name | - | Currently supports L1BalancedCELoss, CTCLoss, AttentionLoss, PSEDiceLoss, EASTLoss and CrossEntropySmooth |
+| name | loss function name | - | Currently supports DBLoss, CTCLoss, AttentionLoss, PSEDiceLoss, EASTLoss and CrossEntropySmooth |
 | pred_seq_len | length of predicted text | 26 | Determined by network architecture |
 | max_label_len | The longest label length | 25 | The value is less than the length of the text predicted by the network |
 | batch_size | single card batch size | 32 | \ |

diff --git a/mindocr/losses/builder.py b/mindocr/losses/builder.py
@@ -1,7 +1,7 @@
 __all__ = ["build_loss"]
 
 supported_losses = [
-    "L1BalancedCELoss",
+    "DBLoss",
     "CTCLoss",
     "AttentionLoss",
     "PSEDiceLoss",
@@ -11,7 +11,7 @@
 ]
 
 from .cls_loss import CrossEntropySmooth
-from .det_loss import EASTLoss, FCELoss, L1BalancedCELoss, PSEDiceLoss
+from .det_loss import DBLoss, EASTLoss, FCELoss, PSEDiceLoss
 from .rec_loss import AttentionLoss, CTCLoss
 
 

diff --git a/mindocr/losses/det_loss.py b/mindocr/losses/det_loss.py
@@ -8,11 +8,10 @@
 import mindspore.numpy as mnp
 from mindspore import Tensor, nn, ops
 
-__all__ = ["L1BalancedCELoss", "PSEDiceLoss", "EASTLoss", "FCELoss"]
+__all__ = ["DBLoss", "PSEDiceLoss", "EASTLoss", "FCELoss"]
 _logger = logging.getLogger(__name__)
 
-
-class L1BalancedCELoss(nn.LossBase):
+class DBLoss(nn.LossBase):
     """
     Apply Balanced CrossEntropy Loss on `binary`, MaskL1Loss on `thresh`, DiceLoss on `thresh_binary` and return
     overall weighted loss.