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

configs/cls/ch_PP-OCRv3/ch_PP-OCRv3_rotnet.yml

+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 100
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_ppocr_v3_rotnet
+  save_epoch_step: 3
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model: null
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: 25
+  infer_mode: false
+  use_space_char: true
+  save_res_path: ./output/rec/predicts_chinese_lite_v2.0.txt
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: L2
+    factor: 1.0e-05
+Architecture:
+  model_type: cls
+  algorithm: CLS
+  Transform: null
+  Backbone:
+    name: MobileNetV1Enhance
+    scale: 0.5
+    last_conv_stride: [1, 2]
+    last_pool_type: avg
+  Neck:
+  Head:
+    name: ClsHead
+    class_dim: 4
+
+Loss:
+  name: ClsLoss
+  main_indicator: acc
+
+PostProcess:
+  name: ClsPostProcess
+
+Metric:
+  name: ClsMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug:
+        use_tia: False
+    - RandAugment:
+    - SSLRotateResize:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys: ["image", "label"]
+  loader:
+    collate_fn: "SSLRotateCollate"
+    shuffle: true
+    batch_size_per_card: 32
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - SSLRotateResize:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys: ["image", "label"]
+  loader:
+    collate_fn: "SSLRotateCollate"
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 64
+    num_workers: 8
+profiler_options: null

configs/rec/rec_svtrnet.yml

+Global:
+  use_gpu: True
+  epoch_num: 20
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/svtr/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations after the 0th iteration
+  eval_batch_step: [0, 2000]
+  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:
+  character_type: en
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_svtr_tiny.txt
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.99
+  epsilon: 0.00000008
+  weight_decay: 0.05
+  no_weight_decay_name: norm pos_embed
+  one_dim_param_no_weight_decay: true
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 2
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR
+  Transform:
+    name: STN_ON
+    tps_inputsize: [32, 64]
+    tps_outputsize: [32, 100]
+    num_control_points: 20
+    tps_margins: [0.05,0.05]
+    stn_activation: none
+  Backbone:
+    name: SVTRNet
+    img_size: [32, 100]
+    out_char_num: 25
+    out_channels: 192
+    patch_merging: 'Conv'
+    embed_dim: [64, 128, 256]
+    depth: [3, 6, 3]
+    num_heads: [2, 4, 8]
+    mixer: ['Local','Local','Local','Local','Local','Local','Global','Global','Global','Global','Global','Global']
+    local_mixer: [[7, 11], [7, 11], [7, 11]]
+    last_stage: True
+    prenorm: false
+  Neck:
+    name: SequenceEncoder
+    encoder_type: reshape
+  Head:
+    name: CTCHead
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          character_dict_path:
+          image_shape: [3, 64, 256]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 512
+    drop_last: True
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/validation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          character_dict_path:
+          image_shape: [3, 64, 256]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 2

deploy/README.md

@@ -22,9 +22,11 @@ PP-OCR has supported muti deployment schemes. Click the link to get the specific
 
 - [Python Inference](../doc/doc_en/inference_ppocr_en.md)
 - [C++ Inference](./cpp_infer/readme.md)
-- [Serving](./pdserving/README.md)
-- [Paddle-Lite](./lite/readme.md)
+- [Serving (Python/C++)](./pdserving/README.md)
+- [Paddle-Lite (ARM CPU/OpenCL ARM GPU/Metal ARM GPU)](./lite/readme.md)
 - [Paddle.js](./paddlejs/README.md)
+- [Jetson Inference]()
+- [XPU Inference]()
 - [Paddle2ONNX](./paddle2onnx/readme.md)
 
 If you need the deployment tutorial of academic algorithm models other than PP-OCR, please directly enter the main page of corresponding algorithms, [entrance](../doc/doc_en/algorithm_overview_en.md)。
\ No newline at end of file

deploy/README_ch.md

@@ -22,9 +22,11 @@ PP-OCR模型已打通多种场景部署方案，点击链接获取具体的使
 
 - [Python 推理](../doc/doc_ch/inference_ppocr.md)
 - [C++ 推理](./cpp_infer/readme_ch.md)
-- [Serving 服务化部署](./pdserving/README_CN.md)
-- [Paddle-Lite 端侧部署](./lite/readme_ch.md)
-- [Paddle.js 服务化部署](./paddlejs/README_ch.md)
+- [Serving 服务化部署（Python/C++）](./pdserving/README_CN.md)
+- [Paddle-Lite 端侧部署（ARM CPU/OpenCL ARM GPU/Metal ARM GPU）](./lite/readme_ch.md)
+- [Paddle.js 部署](./paddlejs/README_ch.md)
+- [Jetson 推理]()
+- [XPU 推理]()
 - [Paddle2ONNX 推理](./paddle2onnx/readme_ch.md)
 
 需要PP-OCR以外的学术算法模型的推理部署，请直接进入相应算法主页面，[入口](../doc/doc_ch/algorithm_overview.md)。
\ No newline at end of file

deploy/slim/quantization/export_model.py

@@ -35,17 +35,7 @@ from ppocr.metrics import build_metric
 import tools.program as program
 from paddleslim.dygraph.quant import QAT
 from ppocr.data import build_dataloader
-
-
-def export_single_model(quanter, model, infer_shape, save_path, logger):
-    quanter.save_quantized_model(
-        model,
-        save_path,
-        input_spec=[
-            paddle.static.InputSpec(
-                shape=[None] + infer_shape, dtype='float32')
-        ])
-    logger.info('inference QAT model is saved to {}'.format(save_path))
+from tools.export_model import export_single_model
 
 
 def main():
@@ -84,17 +74,54 @@ def main():
                                             config['Global'])
 
     # build model
-    # for rec algorithm
     if hasattr(post_process_class, 'character'):
         char_num = len(getattr(post_process_class, 'character'))
         if config['Architecture']["algorithm"] in ["Distillation",
                                                    ]:  # distillation model
             for key in config['Architecture']["Models"]:
-                config['Architecture']["Models"][key]["Head"][
-                    'out_channels'] = char_num
+                if config['Architecture']['Models'][key]['Head'][
+                        'name'] == 'MultiHead':  # for multi head
+                    if config['PostProcess'][
+                            'name'] == 'DistillationSARLabelDecode':
+                        char_num = char_num - 2
+                    # update SARLoss params
+                    assert list(config['Loss']['loss_config_list'][-1].keys())[
+                        0] == 'DistillationSARLoss'
+                    config['Loss']['loss_config_list'][-1][
+                        'DistillationSARLoss']['ignore_index'] = char_num + 1
+                    out_channels_list = {}
+                    out_channels_list['CTCLabelDecode'] = char_num
+                    out_channels_list['SARLabelDecode'] = char_num + 2
+                    config['Architecture']['Models'][key]['Head'][
+                        'out_channels_list'] = out_channels_list
+                else:
+                    config['Architecture']["Models"][key]["Head"][
+                        'out_channels'] = char_num
+        elif config['Architecture']['Head'][
+                'name'] == 'MultiHead':  # for multi head
+            if config['PostProcess']['name'] == 'SARLabelDecode':
+                char_num = char_num - 2
+            # update SARLoss params
+            assert list(config['Loss']['loss_config_list'][1].keys())[
+                0] == 'SARLoss'
+            if config['Loss']['loss_config_list'][1]['SARLoss'] is None:
+                config['Loss']['loss_config_list'][1]['SARLoss'] = {
+                    'ignore_index': char_num + 1
+                }
+            else:
+                config['Loss']['loss_config_list'][1]['SARLoss'][
+                    'ignore_index'] = char_num + 1
+            out_channels_list = {}
+            out_channels_list['CTCLabelDecode'] = char_num
+            out_channels_list['SARLabelDecode'] = char_num + 2
+            config['Architecture']['Head'][
+                'out_channels_list'] = out_channels_list
         else:  # base rec model
             config['Architecture']["Head"]['out_channels'] = char_num
 
+        if config['PostProcess']['name'] == 'SARLabelDecode':  # for SAR model
+            config['Loss']['ignore_index'] = char_num - 1
+
     model = build_model(config['Architecture'])
 
     # get QAT model
@@ -120,21 +147,22 @@ def main():
     for k, v in metric.items():
         logger.info('{}:{}'.format(k, v))
 
-    infer_shape = [3, 32, 100] if model_type == "rec" else [3, 640, 640]
-
     save_path = config["Global"]["save_inference_dir"]
 
     arch_config = config["Architecture"]
+
+    arch_config = config["Architecture"]
+
     if arch_config["algorithm"] in ["Distillation", ]:  # distillation model
+        archs = list(arch_config["Models"].values())
         for idx, name in enumerate(model.model_name_list):
             model.model_list[idx].eval()
             sub_model_save_path = os.path.join(save_path, name, "inference")
-            export_single_model(quanter, model.model_list[idx], infer_shape,
-                                sub_model_save_path, logger)
+            export_single_model(model.model_list[idx], archs[idx],
+                                sub_model_save_path, logger, quanter)
     else:
         save_path = os.path.join(save_path, "inference")
-        model.eval()
-        export_single_model(quanter, model, infer_shape, save_path, logger)
+        export_single_model(model, arch_config, save_path, logger, quanter)
 
 
 if __name__ == "__main__":

deploy/slim/quantization/quant.py

@@ -112,10 +112,48 @@ def main(config, device, logger, vdl_writer):
         if config['Architecture']["algorithm"] in ["Distillation",
                                                    ]:  # distillation model
             for key in config['Architecture']["Models"]:
-                config['Architecture']["Models"][key]["Head"][
-                    'out_channels'] = char_num
+                if config['Architecture']['Models'][key]['Head'][
+                        'name'] == 'MultiHead':  # for multi head
+                    if config['PostProcess'][
+                            'name'] == 'DistillationSARLabelDecode':
+                        char_num = char_num - 2
+                    # update SARLoss params
+                    assert list(config['Loss']['loss_config_list'][-1].keys())[
+                        0] == 'DistillationSARLoss'
+                    config['Loss']['loss_config_list'][-1][
+                        'DistillationSARLoss']['ignore_index'] = char_num + 1
+                    out_channels_list = {}
+                    out_channels_list['CTCLabelDecode'] = char_num
+                    out_channels_list['SARLabelDecode'] = char_num + 2
+                    config['Architecture']['Models'][key]['Head'][
+                        'out_channels_list'] = out_channels_list
+                else:
+                    config['Architecture']["Models"][key]["Head"][
+                        'out_channels'] = char_num
+        elif config['Architecture']['Head'][
+                'name'] == 'MultiHead':  # for multi head
+            if config['PostProcess']['name'] == 'SARLabelDecode':
+                char_num = char_num - 2
+            # update SARLoss params
+            assert list(config['Loss']['loss_config_list'][1].keys())[
+                0] == 'SARLoss'
+            if config['Loss']['loss_config_list'][1]['SARLoss'] is None:
+                config['Loss']['loss_config_list'][1]['SARLoss'] = {
+                    'ignore_index': char_num + 1
+                }
+            else:
+                config['Loss']['loss_config_list'][1]['SARLoss'][
+                    'ignore_index'] = char_num + 1
+            out_channels_list = {}
+            out_channels_list['CTCLabelDecode'] = char_num
+            out_channels_list['SARLabelDecode'] = char_num + 2
+            config['Architecture']['Head'][
+                'out_channels_list'] = out_channels_list
         else:  # base rec model
             config['Architecture']["Head"]['out_channels'] = char_num
+
+        if config['PostProcess']['name'] == 'SARLabelDecode':  # for SAR model
+            config['Loss']['ignore_index'] = char_num - 1
     model = build_model(config['Architecture'])
 
     pre_best_model_dict = dict()
@@ -137,7 +175,7 @@ def main(config, device, logger, vdl_writer):
         config['Optimizer'],
         epochs=config['Global']['epoch_num'],
         step_each_epoch=len(train_dataloader),
-        parameters=model.parameters())
+        model=model)
 
     # resume PACT training process
     if config["Global"]["checkpoints"] is not None:

doc/doc_ch/algorithm_det_db.md

@@ -25,8 +25,8 @@
 
 |模型|骨干网络|配置文件|precision|recall|Hmean|下载链接|
 | --- | --- | --- | --- | --- | --- | --- |
-|DB|ResNet50_vd|configs/det/det_r50_vd_db.yml|86.41%|78.72%|82.38%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)|
-|DB|MobileNetV3|configs/det/det_mv3_db.yml|77.29%|73.08%|75.12%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)|
+|DB|ResNet50_vd|[configs/det/det_r50_vd_db.yml](../../configs/det/det_r50_vd_db.yml)|86.41%|78.72%|82.38%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)|
+|DB|MobileNetV3|[configs/det/det_mv3_db.yml](../../configs/det/det_mv3_db.yml)|77.29%|73.08%|75.12%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)|
 
 
 <a name="2"></a>

doc/doc_ch/dataset/layout_datasets.md

@@ -27,7 +27,7 @@
 
 #### 2、CDLA数据集
 - **数据来源**：https://github.com/buptlihang/CDLA
-- **数据简介**：publaynet数据集的训练集合中包含5000张图像，验证集合中包含1000张图像。总共包含10个类别，分别是： `Text, Title, Figure, Figure caption, Table, Table caption, Header, Footer, Reference, Equation`。部分图像以及标注框可视化如下所示。
+- **数据简介**：CDLA据集的训练集合中包含5000张图像，验证集合中包含1000张图像。总共包含10个类别，分别是： `Text, Title, Figure, Figure caption, Table, Table caption, Header, Footer, Reference, Equation`。部分图像以及标注框可视化如下所示。
 
 <div align="center">
     <img src="../datasets/CDLA_demo/val_0633.jpg" width="500">

doc/doc_ch/ppocr_introduction.md

@@ -17,6 +17,8 @@
 
 PP-OCR是PaddleOCR自研的实用的超轻量OCR系统。在实现[前沿算法](algorithm.md)的基础上，考虑精度与速度的平衡，进行**模型瘦身**和**深度优化**，使其尽可能满足产业落地需求。
 
+#### PP-OCR
+
 PP-OCR是一个两阶段的OCR系统，其中文本检测算法选用[DB](algorithm_det_db.md)，文本识别算法选用[CRNN](algorithm_rec_crnn.md)，并在检测和识别模块之间添加[文本方向分类器](angle_class.md)，以应对不同方向的文本识别。
 
 PP-OCR系统pipeline如下：
@@ -28,9 +30,13 @@ PP-OCR系统pipeline如下：
 
 PP-OCR系统在持续迭代优化，目前已发布PP-OCR和PP-OCRv2两个版本：
 
-[1] PP-OCR从骨干网络选择和调整、预测头部的设计、数据增强、学习率变换策略、正则化参数选择、预训练模型使用以及模型自动裁剪量化8个方面，采用19个有效策略，对各个模块的模型进行效果调优和瘦身(如绿框所示)，最终得到整体大小为3.5M的超轻量中英文OCR和2.8M的英文数字OCR。更多细节请参考PP-OCR技术方案 https://arxiv.org/abs/2009.09941
+PP-OCR从骨干网络选择和调整、预测头部的设计、数据增强、学习率变换策略、正则化参数选择、预训练模型使用以及模型自动裁剪量化8个方面，采用19个有效策略，对各个模块的模型进行效果调优和瘦身(如绿框所示)，最终得到整体大小为3.5M的超轻量中英文OCR和2.8M的英文数字OCR。更多细节请参考PP-OCR技术方案 https://arxiv.org/abs/2009.09941
+
+#### PP-OCRv2
+
+PP-OCRv2在PP-OCR的基础上，进一步在5个方面重点优化，检测模型采用CML协同互学习知识蒸馏策略和CopyPaste数据增广策略；识别模型采用LCNet轻量级骨干网络、UDML 改进知识蒸馏策略和[Enhanced CTC loss](./doc/doc_ch/enhanced_ctc_loss.md)损失函数改进（如上图红框所示），进一步在推理速度和预测效果上取得明显提升。更多细节请参考PP-OCRv2[技术报告](https://arxiv.org/abs/2109.03144)。
 
-[2] PP-OCRv2在PP-OCR的基础上，进一步在5个方面重点优化，检测模型采用CML协同互学习知识蒸馏策略和CopyPaste数据增广策略；识别模型采用LCNet轻量级骨干网络、UDML 改进知识蒸馏策略和[Enhanced CTC loss](./doc/doc_ch/enhanced_ctc_loss.md)损失函数改进（如上图红框所示），进一步在推理速度和预测效果上取得明显提升。更多细节请参考PP-OCRv2[技术报告](https://arxiv.org/abs/2109.03144)。
+#### PP-OCRv3
 
 
 <a name="2"></a>

doc/doc_en/algorithm_det_db_en.md

@@ -25,8 +25,8 @@ On the ICDAR2015 dataset, the text detection result is as follows:
 
 |Model|Backbone|Configuration|Precision|Recall|Hmean|Download|
 | --- | --- | --- | --- | --- | --- | --- |
-|DB|ResNet50_vd|configs/det/det_r50_vd_db.yml|86.41%|78.72%|82.38%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)|
-|DB|MobileNetV3|configs/det/det_mv3_db.yml|77.29%|73.08%|75.12%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)|
+|DB|ResNet50_vd|[configs/det/det_r50_vd_db.yml](../../configs/det/det_r50_vd_db.yml)|86.41%|78.72%|82.38%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)|
+|DB|MobileNetV3|[configs/det/det_mv3_db.yml](../../configs/det/det_mv3_db.yml)|77.29%|73.08%|75.12%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)|
 
 
 <a name="2"></a>

ppocr/data/__init__.py

@@ -72,6 +72,7 @@ def build_dataloader(config, mode, device, logger, seed=None):
         use_shared_memory = loader_config['use_shared_memory']
     else:
         use_shared_memory = True
+
     if mode == "Train":
         # Distribute data to multiple cards
         batch_sampler = DistributedBatchSampler(

ppocr/data/collate_fn.py

@@ -56,3 +56,17 @@ class ListCollator(object):
         for idx in to_tensor_idxs:
             data_dict[idx] = paddle.to_tensor(data_dict[idx])
         return list(data_dict.values())
+
+
+class SSLRotateCollate(object):
+    """
+    bach: [
+        [(4*3xH*W), (4,)]
+        [(4*3xH*W), (4,)]
+        ...
+    ]
+    """
+
+    def __call__(self, batch):
+        output = [np.concatenate(d, axis=0) for d in zip(*batch)]
+        return output

ppocr/data/imaug/__init__.py

@@ -23,7 +23,8 @@ from .random_crop_data import EastRandomCropData, RandomCropImgMask
 from .make_pse_gt import MakePseGt
 
 from .rec_img_aug import RecAug, RecConAug, RecResizeImg, ClsResizeImg, \
-    SRNRecResizeImg, NRTRRecResizeImg, SARRecResizeImg, PRENResizeImg
+    SRNRecResizeImg, NRTRRecResizeImg, SARRecResizeImg, PRENResizeImg, SVTRRecResizeImg
+from .ssl_img_aug import SSLRotateResize
 from .randaugment import RandAugment
 from .copy_paste import CopyPaste
 from .ColorJitter import ColorJitter

ppocr/data/imaug/label_ops.py

@@ -113,14 +113,14 @@ class BaseRecLabelEncode(object):
             dict_character = list(self.character_str)
             self.lower = True
         else:
-            self.character_str = ""
+            self.character_str = []
             with open(character_dict_path, "rb") as fin:
                 lines = fin.readlines()
                 for line in lines:
                     line = line.decode('utf-8').strip("\n").strip("\r\n")
-                    self.character_str += line
+                    self.character_str.append(line)
             if use_space_char:
-                self.character_str += " "
+                self.character_str.append(" ")
             dict_character = list(self.character_str)
         dict_character = self.add_special_char(dict_character)
         self.dict = {}

ppocr/data/imaug/rec_img_aug.py

@@ -16,6 +16,7 @@ import math
 import cv2
 import numpy as np
 import random
+import copy
 from PIL import Image
 from .text_image_aug import tia_perspective, tia_stretch, tia_distort
 
@@ -81,7 +82,7 @@ class ClsResizeImg(object):
 
     def __call__(self, data):
         img = data['image']
-        norm_img = resize_norm_img(img, self.image_shape)
+        norm_img, _ = resize_norm_img(img, self.image_shape)
         data['image'] = norm_img
         return data
 
@@ -206,6 +207,25 @@ class PRENResizeImg(object):
         return data
 
 
+class SVTRRecResizeImg(object):
+    def __init__(self,
+                 image_shape,
+                 infer_mode=False,
+                 character_dict_path='./ppocr/utils/ppocr_keys_v1.txt',
+                 padding=True,
+                 **kwargs):
+        self.image_shape = image_shape
+        self.infer_mode = infer_mode
+        self.character_dict_path = character_dict_path
+        self.padding = padding
+
+    def __call__(self, data):
+        img = data['image']
+        norm_img = resize_norm_img_svtr(img, self.image_shape, self.padding)
+        data['image'] = norm_img
+        return data
+
+
 def resize_norm_img_sar(img, image_shape, width_downsample_ratio=0.25):
     imgC, imgH, imgW_min, imgW_max = image_shape
     h = img.shape[0]
@@ -324,6 +344,58 @@ def resize_norm_img_srn(img, image_shape):
     return np.reshape(img_black, (c, row, col)).astype(np.float32)
 
 
+def resize_norm_img_svtr(img, image_shape, padding=False):
+    imgC, imgH, imgW = image_shape
+    h = img.shape[0]
+    w = img.shape[1]
+    if not padding:
+        if h > 2.0 * w:
+            image = Image.fromarray(img)
+            image1 = image.rotate(90, expand=True)
+            image2 = image.rotate(-90, expand=True)
+            img1 = np.array(image1)
+            img2 = np.array(image2)
+        else:
+            img1 = copy.deepcopy(img)
+            img2 = copy.deepcopy(img)
+
+        resized_image = cv2.resize(
+            img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
+        resized_image1 = cv2.resize(
+            img1, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
+        resized_image2 = cv2.resize(
+            img2, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
+        resized_w = imgW
+    else:
+        ratio = w / float(h)
+        if math.ceil(imgH * ratio) > imgW:
+            resized_w = imgW
+        else:
+            resized_w = int(math.ceil(imgH * ratio))
+        resized_image = cv2.resize(img, (resized_w, imgH))
+    resized_image = resized_image.astype('float32')
+    resized_image1 = resized_image1.astype('float32')
+    resized_image2 = resized_image2.astype('float32')
+    if image_shape[0] == 1:
+        resized_image = resized_image / 255
+        resized_image = resized_image[np.newaxis, :]
+    else:
+        resized_image = resized_image.transpose((2, 0, 1)) / 255
+        resized_image1 = resized_image1.transpose((2, 0, 1)) / 255
+        resized_image2 = resized_image2.transpose((2, 0, 1)) / 255
+    resized_image -= 0.5
+    resized_image /= 0.5
+    resized_image1 -= 0.5
+    resized_image1 /= 0.5
+    resized_image2 -= 0.5
+    resized_image2 /= 0.5
+    padding_im = np.zeros((3, imgC, imgH, imgW), dtype=np.float32)
+    padding_im[0, :, :, 0:resized_w] = resized_image
+    padding_im[1, :, :, 0:resized_w] = resized_image1
+    padding_im[2, :, :, 0:resized_w] = resized_image2
+    return padding_im
+
+
 def srn_other_inputs(image_shape, num_heads, max_text_length):
 
     imgC, imgH, imgW = image_shape

ppocr/data/imaug/ssl_img_aug.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 math
+import cv2
+import numpy as np
+import random
+from PIL import Image
+
+from .rec_img_aug import resize_norm_img
+
+
+class SSLRotateResize(object):
+    def __init__(self,
+                 image_shape,
+                 padding=False,
+                 select_all=True,
+                 mode="train",
+                 **kwargs):
+        self.image_shape = image_shape
+        self.padding = padding
+        self.select_all = select_all
+        self.mode = mode
+
+    def __call__(self, data):
+        img = data["image"]
+
+        data["image_r90"] = cv2.rotate(img, cv2.ROTATE_90_CLOCKWISE)
+        data["image_r180"] = cv2.rotate(data["image_r90"],
+                                        cv2.ROTATE_90_CLOCKWISE)
+        data["image_r270"] = cv2.rotate(data["image_r180"],
+                                        cv2.ROTATE_90_CLOCKWISE)
+
+        images = []
+        for key in ["image", "image_r90", "image_r180", "image_r270"]:
+            images.append(
+                resize_norm_img(
+                    data.pop(key),
+                    image_shape=self.image_shape,
+                    padding=self.padding)[0])
+        data["image"] = np.stack(images, axis=0)
+        data["label"] = np.array(list(range(4)))
+        if not self.select_all:
+            data["image"] = data["image"][0::2]  # just choose 0 and 180
+            data["label"] = data["label"][0:2]  # label needs to be continuous
+        if self.mode == "test":
+            data["image"] = data["image"][0]
+            data["label"] = data["label"][0]
+        return data

ppocr/modeling/backbones/rec_svtrnet.py

@@ -296,47 +296,49 @@ class PatchEmbed(nn.Layer):
         if sub_num == 2:
             self.proj = nn.Sequential(
                 ConvBNLayer(
-                    in_channels,
-                    embed_dim // 2,
-                    3,
-                    2,
-                    1,
+                    in_channels=in_channels,
+                    out_channels=embed_dim // 2,
+                    kernel_size=3,
+                    stride=2,
+                    padding=1,
                     act=nn.GELU,
                     bias_attr=None),
                 ConvBNLayer(
-                    embed_dim // 2,
-                    embed_dim,
-                    3,
-                    2,
-                    1,
+                    in_channels=embed_dim // 2,
+                    out_channels=embed_dim,
+                    kernel_size=3,
+                    stride=2,
+                    padding=1,
                     act=nn.GELU,
                     bias_attr=None))
         if sub_num == 3:
             self.proj = nn.Sequential(
                 ConvBNLayer(
-                    in_channels,
-                    embed_dim // 4,
-                    3,
-                    2,
-                    1,
+                    in_channels=in_channels,
+                    out_channels=embed_dim // 4,
+                    kernel_size=3,
+                    stride=2,
+                    padding=1,
                     act=nn.GELU,
                     bias_attr=None),
                 ConvBNLayer(
-                    embed_dim // 4,
-                    embed_dim // 2,
-                    3,
-                    2,
-                    1,
+                    in_channels=embed_dim // 4,
+                    out_channels=embed_dim // 2,
+                    kernel_size=3,
+                    stride=2,
+                    padding=1,
                     act=nn.GELU,
                     bias_attr=None),
                 ConvBNLayer(
                     embed_dim // 2,
                     embed_dim,
-                    3,
-                    2,
-                    1,
+                    in_channels=embed_dim // 2,
+                    out_channels=embed_dim,
+                    kernel_size=3,
+                    stride=2,
+                    padding=1,
                     act=nn.GELU,
-                    bias_attr=None), )
+                    bias_attr=None))
 
     def forward(self, x):
         B, C, H, W = x.shape
@@ -455,7 +457,7 @@ class SVTRNet(nn.Layer):
                 qkv_bias=qkv_bias,
                 qk_scale=qk_scale,
                 drop=drop_rate,
-                act_layer=nn.Swish,
+                act_layer=eval(act),
                 attn_drop=attn_drop_rate,
                 drop_path=dpr[0:depth[0]][i],
                 norm_layer=norm_layer,

ppocr/modeling/transforms/stn.py

@@ -128,6 +128,8 @@ class STN_ON(nn.Layer):
         self.out_channels = in_channels
 
     def forward(self, image):
+        if len(image.shape)==5:
+            image = image.reshape([0, image.shape[-3], image.shape[-2], image.shape[-1]])
         stn_input = paddle.nn.functional.interpolate(
             image, self.tps_inputsize, mode="bilinear", align_corners=True)
         stn_img_feat, ctrl_points = self.stn_head(stn_input)

ppocr/modeling/transforms/tps_spatial_transformer.py

@@ -138,9 +138,9 @@ class TPSSpatialTransformer(nn.Layer):
         assert source_control_points.shape[2] == 2
         batch_size = paddle.shape(source_control_points)[0]
 
-        self.padding_matrix = paddle.expand(
+        padding_matrix = paddle.expand(
             self.padding_matrix, shape=[batch_size, 3, 2])
-        Y = paddle.concat([source_control_points, self.padding_matrix], 1)
+        Y = paddle.concat([source_control_points, padding_matrix], 1)
         mapping_matrix = paddle.matmul(self.inverse_kernel, Y)
         source_coordinate = paddle.matmul(self.target_coordinate_repr,
                                           mapping_matrix)

ppocr/optimizer/__init__.py

@@ -30,7 +30,7 @@ def build_lr_scheduler(lr_config, epochs, step_each_epoch):
     return lr
 
 
-def build_optimizer(config, epochs, step_each_epoch, parameters):
+def build_optimizer(config, epochs, step_each_epoch, model):
     from . import regularizer, optimizer
     config = copy.deepcopy(config)
     # step1 build lr
@@ -43,6 +43,8 @@ def build_optimizer(config, epochs, step_each_epoch, parameters):
         if not hasattr(regularizer, reg_name):
             reg_name += 'Decay'
         reg = getattr(regularizer, reg_name)(**reg_config)()
+    elif 'weight_decay' in config:
+        reg = config.pop('weight_decay')
     else:
         reg = None
 
@@ -57,4 +59,4 @@ def build_optimizer(config, epochs, step_each_epoch, parameters):
                                            weight_decay=reg,
                                            grad_clip=grad_clip,
                                            **config)
-    return optim(parameters), lr
+    return optim(model), lr

ppocr/optimizer/optimizer.py

@@ -42,13 +42,13 @@ class Momentum(object):
         self.weight_decay = weight_decay
         self.grad_clip = grad_clip
 
-    def __call__(self, parameters):
+    def __call__(self, model):
         opt = optim.Momentum(
             learning_rate=self.learning_rate,
             momentum=self.momentum,
             weight_decay=self.weight_decay,
             grad_clip=self.grad_clip,
-            parameters=parameters)
+            parameters=model.parameters())
         return opt
 
 
@@ -75,7 +75,7 @@ class Adam(object):
         self.name = name
         self.lazy_mode = lazy_mode
 
-    def __call__(self, parameters):
+    def __call__(self, model):
         opt = optim.Adam(
             learning_rate=self.learning_rate,
             beta1=self.beta1,
@@ -85,7 +85,7 @@ class Adam(object):
             grad_clip=self.grad_clip,
             name=self.name,
             lazy_mode=self.lazy_mode,
-            parameters=parameters)
+            parameters=model.parameters())
         return opt
 
 
@@ -117,7 +117,7 @@ class RMSProp(object):
         self.weight_decay = weight_decay
         self.grad_clip = grad_clip
 
-    def __call__(self, parameters):
+    def __call__(self, model):
         opt = optim.RMSProp(
             learning_rate=self.learning_rate,
             momentum=self.momentum,
@@ -125,7 +125,7 @@ class RMSProp(object):
             epsilon=self.epsilon,
             weight_decay=self.weight_decay,
             grad_clip=self.grad_clip,
-            parameters=parameters)
+            parameters=model.parameters())
         return opt
 
 
@@ -148,7 +148,7 @@ class Adadelta(object):
         self.grad_clip = grad_clip
         self.name = name
 
-    def __call__(self, parameters):
+    def __call__(self, model):
         opt = optim.Adadelta(
             learning_rate=self.learning_rate,
             epsilon=self.epsilon,
@@ -156,7 +156,7 @@ class Adadelta(object):
             weight_decay=self.weight_decay,
             grad_clip=self.grad_clip,
             name=self.name,
-            parameters=parameters)
+            parameters=model.parameters())
         return opt
 
 
@@ -165,31 +165,55 @@ class AdamW(object):
                  learning_rate=0.001,
                  beta1=0.9,
                  beta2=0.999,
-                 epsilon=1e-08,
+                 epsilon=1e-8,
                  weight_decay=0.01,
+                 multi_precision=False,
                  grad_clip=None,
+                 no_weight_decay_name=None,
+                 one_dim_param_no_weight_decay=False,
                  name=None,
                  lazy_mode=False,
-                 **kwargs):
+                 **args):
+        super().__init__()
         self.learning_rate = learning_rate
         self.beta1 = beta1
         self.beta2 = beta2
         self.epsilon = epsilon
-        self.learning_rate = learning_rate
+        self.grad_clip = grad_clip
         self.weight_decay = 0.01 if weight_decay is None else weight_decay
         self.grad_clip = grad_clip
         self.name = name
         self.lazy_mode = lazy_mode
-
-    def __call__(self, parameters):
+        self.multi_precision = multi_precision
+        self.no_weight_decay_name_list = no_weight_decay_name.split(
+        ) if no_weight_decay_name else []
+        self.one_dim_param_no_weight_decay = one_dim_param_no_weight_decay
+
+    def __call__(self, model):
+        parameters = model.parameters()
+
+        self.no_weight_decay_param_name_list = [
+            p.name for n, p in model.named_parameters() if any(nd in n for nd in self.no_weight_decay_name_list)
+        ]
+
+        if self.one_dim_param_no_weight_decay:
+            self.no_weight_decay_param_name_list += [
+                p.name  for n, p in model.named_parameters() if len(p.shape) == 1
+            ]
+        
         opt = optim.AdamW(
             learning_rate=self.learning_rate,
             beta1=self.beta1,
             beta2=self.beta2,
             epsilon=self.epsilon,
+            parameters=parameters,
             weight_decay=self.weight_decay,
+            multi_precision=self.multi_precision,
             grad_clip=self.grad_clip,
             name=self.name,
             lazy_mode=self.lazy_mode,
-            parameters=parameters)
+            apply_decay_param_fun=self._apply_decay_param_fun)
         return opt
+
+    def _apply_decay_param_fun(self, name):
+        return name not in self.no_weight_decay_param_name_list
\ No newline at end of file

ppocr/postprocess/__init__.py

@@ -27,7 +27,7 @@ from .sast_postprocess import SASTPostProcess
 from .fce_postprocess import FCEPostProcess
 from .rec_postprocess import CTCLabelDecode, AttnLabelDecode, SRNLabelDecode, \
     DistillationCTCLabelDecode, TableLabelDecode, NRTRLabelDecode, SARLabelDecode, \
-    SEEDLabelDecode, PRENLabelDecode
+    SEEDLabelDecode, PRENLabelDecode, SVTRLabelDecode
 from .cls_postprocess import ClsPostProcess
 from .pg_postprocess import PGPostProcess
 from .vqa_token_ser_layoutlm_postprocess import VQASerTokenLayoutLMPostProcess
@@ -42,7 +42,7 @@ def build_post_process(config, global_config=None):
         'DistillationDBPostProcess', 'NRTRLabelDecode', 'SARLabelDecode',
         'SEEDLabelDecode', 'VQASerTokenLayoutLMPostProcess',
         'VQAReTokenLayoutLMPostProcess', 'PRENLabelDecode',
-        'DistillationSARLabelDecode'
+        'DistillationSARLabelDecode', 'SVTRLabelDecode'
     ]
 
     if config['name'] == 'PSEPostProcess':

ppocr/postprocess/cls_postprocess.py

@@ -17,17 +17,26 @@ import paddle
 class ClsPostProcess(object):
     """ Convert between text-label and text-index """
 
-    def __init__(self, label_list, **kwargs):
+    def __init__(self, label_list=None, key=None, **kwargs):
         super(ClsPostProcess, self).__init__()
         self.label_list = label_list
+        self.key = key
 
     def __call__(self, preds, label=None, *args, **kwargs):
+        if self.key is not None:
+            preds = preds[self.key]
+
+        label_list = self.label_list
+        if label_list is None:
+            label_list = {idx: idx for idx in range(preds.shape[-1])}
+
         if isinstance(preds, paddle.Tensor):
             preds = preds.numpy()
+
         pred_idxs = preds.argmax(axis=1)
-        decode_out = [(self.label_list[idx], preds[i, idx])
+        decode_out = [(label_list[idx], preds[i, idx])
                       for i, idx in enumerate(pred_idxs)]
         if label is None:
             return decode_out
-        label = [(self.label_list[idx], 1.0) for idx in label]
+        label = [(label_list[idx], 1.0) for idx in label]
         return decode_out, label

ppocr/postprocess/rec_postprocess.py

@@ -752,3 +752,40 @@ class PRENLabelDecode(BaseRecLabelDecode):
             return text
         label = self.decode(label)
         return text, label
+
+
+class SVTRLabelDecode(BaseRecLabelDecode):
+    """ Convert between text-label and text-index """
+
+    def __init__(self, character_dict_path=None, use_space_char=False,
+                 **kwargs):
+        super(SVTRLabelDecode, self).__init__(character_dict_path,
+                                             use_space_char)
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        if isinstance(preds, tuple):
+            preds = preds[-1]
+        if isinstance(preds, paddle.Tensor):
+            preds = preds.numpy()
+        preds_idx = preds.argmax(axis=-1)
+        preds_prob = preds.max(axis=-1)
+
+        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=True)
+        return_text = []
+        for i in range(0, len(text), 3):
+            text0 = text[i]
+            text1 = text[i + 1]
+            text2 = text[i + 2]
+
+            text_pred = [text0[0], text1[0], text2[0]]
+            text_prob = [text0[1], text1[1], text2[1]]
+            id_max = text_prob.index(max(text_prob))
+            return_text.append((text_pred[id_max], text_prob[id_max]))
+        if label is None:
+            return return_text
+        label = self.decode(label)
+        return return_text, label
+
+    def add_special_char(self, dict_character):
+        dict_character = ['blank'] + dict_character
+        return dict_character
\ No newline at end of file

ppocr/utils/dict/ka_dict.txt

@@ -21,7 +21,7 @@ l
 8
 .
 j
-p	
+p
 ಗ
 ು
 ಣ

ppocr/utils/dict/ta_dict.txt

@@ -22,7 +22,7 @@ l
 8
 .
 j
-p	
+p
 ப
 ூ
 த

test_tipc/configs/ch_PP-OCRv2_det/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

+===========================train_params===========================
+model_name:ch_PPOCRv2_det
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=500
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o 
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o 
+quant_export:null
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/ch_PP-OCRv2_det_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1
+--use_tensorrt:False|True
+--precision:fp32|fp16|int8
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]

test_tipc/configs/ch_PP-OCRv2_det_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

+===========================train_params===========================
+model_name:ch_PPOCRv2_det_PACT
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=500
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:pact_train
+norm_train:null
+pact_train:deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null
+quant_export:deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o 
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/ch_PP-OCRv2_det_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1
+--use_tensorrt:False|True
+--precision:fp32|fp16|int8
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]

test_tipc/configs/ch_PP-OCRv2_rec/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

+===========================train_params===========================
+model_name:PPOCRv2_ocr_rec
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml -o 
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/ch_PP-OCRv2_rec_infer
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py
+--use_gpu:True|False
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1|6
+--use_tensorrt:False|True
+--precision:fp32|int8
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,320]}]

test_tipc/configs/ch_PP-OCRv2_rec_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

+===========================train_params===========================
+model_name:ch_PPOCRv2_rec_PACT
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:pact_train
+norm_train:null
+pact_train:deploy/slim/quantization/quant.py -c test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml -o 
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null
+quant_export:deploy/slim/quantization/export_model.py -c test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml -o 
+fpgm_export: null
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/ch_PP-OCRv2_rec_slim_quant_infer
+infer_export:null
+infer_quant:True
+inference:tools/infer/predict_rec.py
+--use_gpu:True|False
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1|6
+--use_tensorrt:False|True
+--precision:fp32|int8
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,320]}]

test_tipc/configs/ch_ppocr_mobile_v2.0_det/det_mv3_db.yml

-Global:
-  use_gpu: false
-  epoch_num: 5
-  log_smooth_window: 20
-  print_batch_step: 1
-  save_model_dir: ./output/db_mv3/
-  save_epoch_step: 1200
-  # evaluation is run every 2000 iterations
-  eval_batch_step: [0, 400]
-  cal_metric_during_train: False
-  pretrained_model: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
-  checkpoints:
-  save_inference_dir:
-  use_visualdl: False
-  infer_img: doc/imgs_en/img_10.jpg
-  save_res_path: ./output/det_db/predicts_db.txt
-
-Architecture:
-  model_type: det
-  algorithm: DB
-  Transform:
-  Backbone:
-    name: MobileNetV3
-    scale: 0.5
-    model_name: large
-    disable_se: False
-  Neck:
-    name: DBFPN
-    out_channels: 256
-  Head:
-    name: DBHead
-    k: 50
-
-Loss:
-  name: DBLoss
-  balance_loss: true
-  main_loss_type: DiceLoss
-  alpha: 5 
-  beta: 10 
-  ohem_ratio: 3
-
-Optimizer:
-  name: Adam #Momentum
-  #momentum: 0.9
-  beta1: 0.9
-  beta2: 0.999
-  lr:
-    learning_rate: 0.001
-  regularizer:
-    name: 'L2'
-    factor: 0
-
-PostProcess:
-  name: DBPostProcess
-  thresh: 0.3
-  box_thresh: 0.6
-  max_candidates: 1000
-  unclip_ratio: 1.5
-
-Metric:
-  name: DetMetric
-  main_indicator: hmean
-
-Train:
-  dataset:
-    name: SimpleDataSet
-    data_dir: ./train_data/icdar2015/text_localization/
-    label_file_list:
-      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
-    ratio_list: [1.0]
-    transforms:
-      - DecodeImage: # load image
-          img_mode: BGR
-          channel_first: False
-      - DetLabelEncode: # Class handling label
-      - Resize:
-          size: [640, 640]
-      - MakeBorderMap:
-          shrink_ratio: 0.4
-          thresh_min: 0.3
-          thresh_max: 0.7
-      - MakeShrinkMap:
-          shrink_ratio: 0.4
-          min_text_size: 8
-      - NormalizeImage:
-          scale: 1./255.
-          mean: [0.485, 0.456, 0.406]
-          std: [0.229, 0.224, 0.225]
-          order: 'hwc'
-      - ToCHWImage:
-      - KeepKeys:
-          keep_keys: ['image', 'threshold_map', 'threshold_mask', 'shrink_map', 'shrink_mask'] # the order of the dataloader list
-  loader:
-    shuffle: False
-    drop_last: False
-    batch_size_per_card: 1
-    num_workers: 0
-    use_shared_memory: False
-
-Eval:
-  dataset:
-    name: SimpleDataSet
-    data_dir: ./train_data/icdar2015/text_localization/
-    label_file_list:
-      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
-    transforms:
-      - DecodeImage: # load image
-          img_mode: BGR
-          channel_first: False
-      - DetLabelEncode: # Class handling label
-      - DetResizeForTest:
-          image_shape: [736, 1280]
-      - NormalizeImage:
-          scale: 1./255.
-          mean: [0.485, 0.456, 0.406]
-          std: [0.229, 0.224, 0.225]
-          order: 'hwc'
-      - ToCHWImage:
-      - KeepKeys:
-          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
-  loader:
-    shuffle: False
-    drop_last: False
-    batch_size_per_card: 1 # must be 1
-    num_workers: 0
-    use_shared_memory: False

test_tipc/configs/ch_ppocr_mobile_v2.0_det/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

 ===========================train_params===========================
-model_name:ocr_det
+model_name:ch_ppocr_mobile_v2.0_det
 python:python3.7
 gpu_list:0|0,1
 Global.use_gpu:True|True
 Global.auto_cast:amp
-Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=300
+Global.epoch_num:lite_train_lite_infer=100|whole_train_whole_infer=300
 Global.save_model_dir:./output/
 Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
 Global.pretrained_model:null
@@ -12,10 +12,10 @@ train_model_name:latest
 train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
 null:null
 ##
-trainer:norm_train|pact_train|fpgm_train
-norm_train:tools/train.py -c test_tipc/configs/ppocr_det_mobile/det_mv3_db.yml -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained
-pact_train:deploy/slim/quantization/quant.py -c test_tipc/configs/ppocr_det_mobile/det_mv3_db.yml -o
-fpgm_train:deploy/slim/prune/sensitivity_anal.py -c test_tipc/configs/ppocr_det_mobile/det_mv3_db.yml -o Global.pretrained_model=./pretrain_models/det_mv3_db_v2.0_train/best_accuracy
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained
+pact_train:null
+fpgm_train:null
 distill_train:null
 null:null
 null:null
@@ -26,10 +26,10 @@ null:null
 ##
 ===========================infer_params===========================
 Global.save_inference_dir:./output/
-Global.pretrained_model:
-norm_export:tools/export_model.py -c test_tipc/configs/ppocr_det_mobile/det_mv3_db.yml -o 
-quant_export:deploy/slim/quantization/export_model.py -c test_tipc/configs/ppocr_det_mobile/det_mv3_db.yml -o 
-fpgm_export:deploy/slim/prune/export_prune_model.py -c test_tipc/configs/ppocr_det_mobile/det_mv3_db.yml -o 
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o 
+quant_export:null
+fpgm_export:null
 distill_export:null
 export1:null
 export2:null
@@ -49,3 +49,5 @@ inference:tools/infer/predict_det.py
 null:null
 --benchmark:True
 null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
\ No newline at end of file

test_tipc/configs/ch_ppocr_mobile_V2.0_det_FPGM/train_infer_python.txt → test_tipc/configs/ch_ppocr_mobile_v2.0_det_FPGM/train_infer_python.txt

 ===========================train_params===========================
-model_name:ocr_det
+model_name:ch_ppocr_mobile_v2.0_det_FPGM
 python:python3.7
 gpu_list:0|0,1
 Global.use_gpu:True|True

test_tipc/configs/ch_ppocr_mobile_v2.0_det_FPGM/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

+===========================train_params===========================
+model_name:ch_ppocr_mobile_v2.0_det_FPGM
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=5|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:fpgm_train
+norm_train:null
+pact_train:null
+fpgm_train:deploy/slim/prune/sensitivity_anal.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrained_model=./pretrain_models/det_mv3_db_v2.0_train/best_accuracy
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null
+quant_export:null
+fpgm_export:deploy/slim/prune/export_prune_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o 
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:null
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1
+--use_tensorrt:False|True
+--precision:fp32|fp16|int8
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
\ No newline at end of file

test_tipc/configs/ch_ppocr_mobile_v2.0_det_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

+===========================train_params===========================
+model_name:ch_ppocr_mobile_v2.0_det_PACT
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=20|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:pact_train
+norm_train:null
+pact_train:deploy/slim/quantization/quant.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null
+quant_export:deploy/slim/quantization/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o 
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:./inference/ch_ppocr_mobile_v2.0_det_prune_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1
+--use_tensorrt:False|True
+--precision:fp32|fp16|int8
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]

test_tipc/configs/ch_ppocr_mobile_v2.0_rec/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

+===========================train_params===========================
+model_name:ch_ppocr_mobile_v2.0_rec
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/rec/rec_icdar15_train.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c configs/rec/rec_icdar15_train.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/rec/rec_icdar15_train.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/ch_ppocr_mobile_v2.0_rec_train/best_accuracy
+infer_export:tools/export_model.py -c configs/rec/rec_icdar15_train.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py
+--use_gpu:True|False
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1|6
+--use_tensorrt:True|False
+--precision:fp32|int8
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,100]}]

test_tipc/configs/ch_ppocr_mobile_v2.0_rec_FPGM/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

+===========================train_params===========================
+model_name:ch_ppocr_mobile_v2.0_rec_FPGM
+python:python3.7
+gpu_list:0
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/ic15_data/test/word_1.png
+null:null
+##
+trainer:fpgm_train
+norm_train:null
+pact_train:null
+fpgm_train:deploy/slim/prune/sensitivity_anal.py -c test_tipc/configs/ch_ppocr_mobile_v2.0_rec_FPGM/rec_chinese_lite_train_v2.0.yml -o Global.pretrained_model=./pretrain_models/ch_ppocr_mobile_v2.0_rec_train/best_accuracy
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null
+quant_export:null
+fpgm_export:deploy/slim/prune/export_prune_model.py -c test_tipc/configs/ch_ppocr_mobile_v2.0_rec_FPGM/rec_chinese_lite_train_v2.0.yml -o 
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:null
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py
+--use_gpu:True|False
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1
+--use_tensorrt:False|True
+--precision:fp32|int8
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,320]}]
\ No newline at end of file

test_tipc/configs/ch_ppocr_mobile_v2.0_rec_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

+===========================train_params===========================
+model_name:ch_ppocr_mobile_v2.0_rec_PACT
+python:python3.7
+gpu_list:0
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.checkpoints:null
+train_model_name:latest
+train_infer_img_dir:./train_data/ic15_data/test/word_1.png
+null:null
+##
+trainer:pact_train
+norm_train:null
+pact_train:deploy/slim/quantization/quant.py -c test_tipc/configs/ch_ppocr_mobile_v2.0_rec_PACT/rec_chinese_lite_train_v2.0.yml -o
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null
+quant_export:deploy/slim/quantization/export_model.py -c test_tipc/configs/ch_ppocr_mobile_v2.0_rec_PACT/rec_chinese_lite_train_v2.0.yml -o 
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+infer_model:./inference/ch_ppocr_mobile_v2.0_rec_slim_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --rec_image_shape="3,32,100"
+--use_gpu:True|False
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1|6
+--use_tensorrt:False|True
+--precision:fp32|int8
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,320]}]

test_tipc/configs/ch_ppocr_server_v2.0_det/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

+===========================train_params===========================
+model_name:ch_ppocr_server_v2.0_det
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_lite_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/ch_ppocr_server_v2.0_det/det_r50_vd_db.yml -o 
+quant_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:tools/eval.py -c test_tipc/configs/ch_ppocr_server_v2.0_det/det_r50_vd_db.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/ch_ppocr_server_v2.0_det/det_r50_vd_db.yml -o 
+quant_export:null 
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/ch_ppocr_server_v2.0_det_train/best_accuracy
+infer_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_res18_db_v2.0.yml -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1
+--use_tensorrt:False|True
+--precision:fp32|fp16|int8
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--save_log_path:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
\ No newline at end of file

test_tipc/configs/ch_ppocr_server_v2.0_rec/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

+===========================train_params===========================
+model_name:ch_ppocr_server_v2.0_rec
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=5|whole_train_whole_infer=100
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/ch_ppocr_server_v2.0_rec/rec_icdar15_train.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/ch_ppocr_server_v2.0_rec/rec_icdar15_train.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/ch_ppocr_server_v2.0_rec/rec_icdar15_train.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/ch_ppocr_server_v2.0_rec_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/ch_ppocr_server_v2.0_rec/rec_icdar15_train.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py
+--use_gpu:True|False
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1|6
+--use_tensorrt:True|False
+--precision:fp32|int8
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,100]}]

tools/eval.py

@@ -74,9 +74,11 @@ def main():
 
     model = build_model(config['Architecture'])
     extra_input_models = ["SRN", "NRTR", "SAR", "SEED", "SVTR"]
+    extra_input = False
     if config['Architecture']['algorithm'] == 'Distillation':
-        extra_input = config['Architecture']['Models']['Teacher'][
-            'algorithm'] in extra_input_models
+        for key in config['Architecture']["Models"]:
+            extra_input = extra_input or config['Architecture']['Models'][key][
+                'algorithm'] in extra_input_models
     else:
         extra_input = config['Architecture']['algorithm'] in extra_input_models
     if "model_type" in config['Architecture'].keys():

tools/export_model.py

@@ -31,7 +31,7 @@ from ppocr.utils.logging import get_logger
 from tools.program import load_config, merge_config, ArgsParser
 
 
-def export_single_model(model, arch_config, save_path, logger):
+def export_single_model(model, arch_config, save_path, logger, quanter=None):
     if arch_config["algorithm"] == "SRN":
         max_text_length = arch_config["Head"]["max_text_length"]
         other_shape = [
@@ -61,6 +61,11 @@ def export_single_model(model, arch_config, save_path, logger):
                 paddle.static.InputSpec(
                     shape=[None, 3, 48, -1], dtype="float32"),
             ]
+        else:
+            other_shape = [
+                paddle.static.InputSpec(
+                    shape=[None, 3, 64, 256], dtype="float32"),
+            ]
         model = to_static(model, input_spec=other_shape)
     elif arch_config["algorithm"] == "PREN":
         other_shape = [
@@ -90,7 +95,10 @@ def export_single_model(model, arch_config, save_path, logger):
                     shape=[None] + infer_shape, dtype="float32")
             ])
 
-    paddle.jit.save(model, save_path)
+    if quanter is None:
+        paddle.jit.save(model, save_path)
+    else:
+        quanter.save_quantized_model(model, save_path)
     logger.info("inference model is saved to {}".format(save_path))
     return
 
@@ -120,7 +128,6 @@ def main():
                         char_num = char_num - 2
                     out_channels_list['CTCLabelDecode'] = char_num
                     out_channels_list['SARLabelDecode'] = char_num + 2
-                    loss_list = config['Loss']['loss_config_list']
                     config['Architecture']['Models'][key]['Head'][
                         'out_channels_list'] = out_channels_list
                 else:

tools/infer/predict_rec.py

@@ -131,6 +131,17 @@ class TextRecognizer(object):
         padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
         padding_im[:, :, 0:resized_w] = resized_image
         return padding_im
+    
+    def resize_norm_img_svtr(self, img, image_shape):
+
+        imgC, imgH, imgW = image_shape
+        resized_image = cv2.resize(
+            img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
+        resized_image = resized_image.astype('float32')
+        resized_image = resized_image.transpose((2, 0, 1)) / 255
+        resized_image -= 0.5
+        resized_image /= 0.5
+        return resized_image
 
     def resize_norm_img_srn(self, img, image_shape):
         imgC, imgH, imgW = image_shape
@@ -263,12 +274,8 @@ class TextRecognizer(object):
                 wh_ratio = w * 1.0 / h
                 max_wh_ratio = max(max_wh_ratio, wh_ratio)
             for ino in range(beg_img_no, end_img_no):
-                if self.rec_algorithm != "SRN" and self.rec_algorithm != "SAR":
-                    norm_img = self.resize_norm_img(img_list[indices[ino]],
-                                                    max_wh_ratio)
-                    norm_img = norm_img[np.newaxis, :]
-                    norm_img_batch.append(norm_img)
-                elif self.rec_algorithm == "SAR":
+               
+                if self.rec_algorithm == "SAR":
                     norm_img, _, _, valid_ratio = self.resize_norm_img_sar(
                         img_list[indices[ino]], self.rec_image_shape)
                     norm_img = norm_img[np.newaxis, :]
@@ -276,7 +283,7 @@ class TextRecognizer(object):
                     valid_ratios = []
                     valid_ratios.append(valid_ratio)
                     norm_img_batch.append(norm_img)
-                else:
+                elif self.rec_algorithm == "SRN":
                     norm_img = self.process_image_srn(
                         img_list[indices[ino]], self.rec_image_shape, 8, 25)
                     encoder_word_pos_list = []
@@ -288,6 +295,16 @@ class TextRecognizer(object):
                     gsrm_slf_attn_bias1_list.append(norm_img[3])
                     gsrm_slf_attn_bias2_list.append(norm_img[4])
                     norm_img_batch.append(norm_img[0])
+                elif self.rec_algorithm == "SVTR":
+                    norm_img = self.resize_norm_img_svtr(
+                        img_list[indices[ino]], self.rec_image_shape)
+                    norm_img = norm_img[np.newaxis, :]
+                    norm_img_batch.append(norm_img)
+                else:
+                    norm_img = self.resize_norm_img(img_list[indices[ino]],
+                                                    max_wh_ratio)
+                    norm_img = norm_img[np.newaxis, :]
+                    norm_img_batch.append(norm_img)
             norm_img_batch = np.concatenate(norm_img_batch)
             norm_img_batch = norm_img_batch.copy()
             if self.benchmark:

tools/infer/utility.py

@@ -271,9 +271,10 @@ def create_predictor(args, mode, logger):
             elif mode == "rec":
                 if args.rec_algorithm != "CRNN":
                     use_dynamic_shape = False
-                min_input_shape = {"x": [1, 3, 32, 10]}
-                max_input_shape = {"x": [args.rec_batch_num, 3, 32, 1536]}
-                opt_input_shape = {"x": [args.rec_batch_num, 3, 32, 320]}
+                imgH = int(args.rec_image_shape.split(',')[-2])
+                min_input_shape = {"x": [1, 3, imgH, 10]}
+                max_input_shape = {"x": [args.rec_batch_num, 3, imgH, 1536]}
+                opt_input_shape = {"x": [args.rec_batch_num, 3, imgH, 320]}
             elif mode == "cls":
                 min_input_shape = {"x": [1, 3, 48, 10]}
                 max_input_shape = {"x": [args.rec_batch_num, 3, 48, 1024]}
@@ -300,8 +301,8 @@ def create_predictor(args, mode, logger):
         # enable memory optim
         config.enable_memory_optim()
         config.disable_glog_info()
-
         config.delete_pass("conv_transpose_eltwiseadd_bn_fuse_pass")
+        config.delete_pass("matmul_transpose_reshape_fuse_pass")
         if mode == 'table':
             config.delete_pass("fc_fuse_pass")  # not supported for table
         config.switch_use_feed_fetch_ops(False)

tools/infer_cls.py

@@ -57,6 +57,8 @@ def main():
             continue
         elif op_name == 'KeepKeys':
             op[op_name]['keep_keys'] = ['image']
+        elif op_name == "SSLRotateResize":
+            op[op_name]["mode"] = "test"
         transforms.append(op)
     global_config['infer_mode'] = True
     ops = create_operators(transforms, global_config)

tools/program.py

@@ -202,9 +202,11 @@ def train(config,
 
     use_srn = config['Architecture']['algorithm'] == "SRN"
     extra_input_models = ["SRN", "NRTR", "SAR", "SEED", "SVTR"]
+    extra_input = False
     if config['Architecture']['algorithm'] == 'Distillation':
-        extra_input = config['Architecture']['Models']['Teacher'][
-            'algorithm'] in extra_input_models
+        for key in config['Architecture']["Models"]:
+            extra_input = extra_input or config['Architecture']['Models'][key][
+                'algorithm'] in extra_input_models
     else:
         extra_input = config['Architecture']['algorithm'] in extra_input_models
     try:

tools/train.py

@@ -129,7 +129,7 @@ def main(config, device, logger, vdl_writer):
         config['Optimizer'],
         epochs=config['Global']['epoch_num'],
         step_each_epoch=len(train_dataloader),
-        parameters=model.parameters())
+        model=model)
 
     # build metric
     eval_class = build_metric(config['Metric'])