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

deploy/pdserving/ocr_reader.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.
+
+import cv2
+import copy
+import numpy as np
+import math
+import re
+import sys
+import argparse
+import string
+from copy import deepcopy
+import paddle
+
+
+class DetResizeForTest(object):
+    def __init__(self, **kwargs):
+        super(DetResizeForTest, self).__init__()
+        self.resize_type = 0
+        if 'image_shape' in kwargs:
+            self.image_shape = kwargs['image_shape']
+            self.resize_type = 1
+        elif 'limit_side_len' in kwargs:
+            self.limit_side_len = kwargs['limit_side_len']
+            self.limit_type = kwargs.get('limit_type', 'min')
+        elif 'resize_long' in kwargs:
+            self.resize_type = 2
+            self.resize_long = kwargs.get('resize_long', 960)
+        else:
+            self.limit_side_len = 736
+            self.limit_type = 'min'
+
+    def __call__(self, data):
+        img = deepcopy(data)
+        src_h, src_w, _ = img.shape
+
+        if self.resize_type == 0:
+            img, [ratio_h, ratio_w] = self.resize_image_type0(img)
+        elif self.resize_type == 2:
+            img, [ratio_h, ratio_w] = self.resize_image_type2(img)
+        else:
+            img, [ratio_h, ratio_w] = self.resize_image_type1(img)
+
+        return img
+
+    def resize_image_type1(self, img):
+        resize_h, resize_w = self.image_shape
+        ori_h, ori_w = img.shape[:2]  # (h, w, c)
+        ratio_h = float(resize_h) / ori_h
+        ratio_w = float(resize_w) / ori_w
+        img = cv2.resize(img, (int(resize_w), int(resize_h)))
+        return img, [ratio_h, ratio_w]
+
+    def resize_image_type0(self, img):
+        """
+        resize image to a size multiple of 32 which is required by the network
+        args:
+            img(array): array with shape [h, w, c]
+        return(tuple):
+            img, (ratio_h, ratio_w)
+        """
+        limit_side_len = self.limit_side_len
+        h, w, _ = img.shape
+
+        # limit the max side
+        if self.limit_type == 'max':
+            if max(h, w) > limit_side_len:
+                if h > w:
+                    ratio = float(limit_side_len) / h
+                else:
+                    ratio = float(limit_side_len) / w
+            else:
+                ratio = 1.
+        else:
+            if min(h, w) < limit_side_len:
+                if h < w:
+                    ratio = float(limit_side_len) / h
+                else:
+                    ratio = float(limit_side_len) / w
+            else:
+                ratio = 1.
+        resize_h = int(h * ratio)
+        resize_w = int(w * ratio)
+
+        resize_h = int(round(resize_h / 32) * 32)
+        resize_w = int(round(resize_w / 32) * 32)
+
+        try:
+            if int(resize_w) <= 0 or int(resize_h) <= 0:
+                return None, (None, None)
+            img = cv2.resize(img, (int(resize_w), int(resize_h)))
+        except:
+            print(img.shape, resize_w, resize_h)
+            sys.exit(0)
+        ratio_h = resize_h / float(h)
+        ratio_w = resize_w / float(w)
+        # return img, np.array([h, w])
+        return img, [ratio_h, ratio_w]
+
+    def resize_image_type2(self, img):
+        h, w, _ = img.shape
+
+        resize_w = w
+        resize_h = h
+
+        # Fix the longer side
+        if resize_h > resize_w:
+            ratio = float(self.resize_long) / resize_h
+        else:
+            ratio = float(self.resize_long) / resize_w
+
+        resize_h = int(resize_h * ratio)
+        resize_w = int(resize_w * ratio)
+
+        max_stride = 128
+        resize_h = (resize_h + max_stride - 1) // max_stride * max_stride
+        resize_w = (resize_w + max_stride - 1) // max_stride * max_stride
+        img = cv2.resize(img, (int(resize_w), int(resize_h)))
+        ratio_h = resize_h / float(h)
+        ratio_w = resize_w / float(w)
+
+        return img, [ratio_h, ratio_w]
+
+
+class BaseRecLabelDecode(object):
+    """ Convert between text-label and text-index """
+
+    def __init__(self, config):
+        support_character_type = [
+            'ch', 'en', 'EN_symbol', 'french', 'german', 'japan', 'korean',
+            'it', 'xi', 'pu', 'ru', 'ar', 'ta', 'ug', 'fa', 'ur', 'rs', 'oc',
+            'rsc', 'bg', 'uk', 'be', 'te', 'ka', 'chinese_cht', 'hi', 'mr',
+            'ne', 'EN'
+        ]
+        character_type = config['character_type']
+        character_dict_path = config['character_dict_path']
+        use_space_char = True
+        assert character_type in support_character_type, "Only {} are supported now but get {}".format(
+            support_character_type, character_type)
+
+        self.beg_str = "sos"
+        self.end_str = "eos"
+
+        if character_type == "en":
+            self.character_str = "0123456789abcdefghijklmnopqrstuvwxyz"
+            dict_character = list(self.character_str)
+        elif character_type == "EN_symbol":
+            # same with ASTER setting (use 94 char).
+            self.character_str = string.printable[:-6]
+            dict_character = list(self.character_str)
+        elif character_type in support_character_type:
+            self.character_str = ""
+            assert character_dict_path is not None, "character_dict_path should not be None when character_type is {}".format(
+                character_type)
+            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
+            if use_space_char:
+                self.character_str += " "
+            dict_character = list(self.character_str)
+
+        else:
+            raise NotImplementedError
+        self.character_type = character_type
+        dict_character = self.add_special_char(dict_character)
+        self.dict = {}
+        for i, char in enumerate(dict_character):
+            self.dict[char] = i
+        self.character = dict_character
+
+    def add_special_char(self, dict_character):
+        return dict_character
+
+    def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
+        """ convert text-index into text-label. """
+        result_list = []
+        ignored_tokens = self.get_ignored_tokens()
+        batch_size = len(text_index)
+        for batch_idx in range(batch_size):
+            char_list = []
+            conf_list = []
+            for idx in range(len(text_index[batch_idx])):
+                if text_index[batch_idx][idx] in ignored_tokens:
+                    continue
+                if is_remove_duplicate:
+                    # only for predict
+                    if idx > 0 and text_index[batch_idx][idx - 1] == text_index[
+                            batch_idx][idx]:
+                        continue
+                char_list.append(self.character[int(text_index[batch_idx][
+                    idx])])
+                if text_prob is not None:
+                    conf_list.append(text_prob[batch_idx][idx])
+                else:
+                    conf_list.append(1)
+            text = ''.join(char_list)
+            result_list.append((text, np.mean(conf_list)))
+        return result_list
+
+    def get_ignored_tokens(self):
+        return [0]  # for ctc blank
+
+
+class CTCLabelDecode(BaseRecLabelDecode):
+    """ Convert between text-label and text-index """
+
+    def __init__(
+            self,
+            config,
+            #character_dict_path=None,
+            #character_type='ch',
+            #use_space_char=False,
+            **kwargs):
+        super(CTCLabelDecode, self).__init__(config)
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        if isinstance(preds, paddle.Tensor):
+            preds = preds.numpy()
+        preds_idx = preds.argmax(axis=2)
+        preds_prob = preds.max(axis=2)
+        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=True)
+        if label is None:
+            return text
+        label = self.decode(label)
+        return text, label
+
+    def add_special_char(self, dict_character):
+        dict_character = ['blank'] + dict_character
+        return dict_character
+
+
+class CharacterOps(object):
+    """ Convert between text-label and text-index """
+
+    def __init__(self, config):
+        self.character_type = config['character_type']
+        self.loss_type = config['loss_type']
+        if self.character_type == "en":
+            self.character_str = "0123456789abcdefghijklmnopqrstuvwxyz"
+            dict_character = list(self.character_str)
+        elif self.character_type == "ch":
+            character_dict_path = config['character_dict_path']
+            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
+            dict_character = list(self.character_str)
+        elif self.character_type == "en_sensitive":
+            # same with ASTER setting (use 94 char).
+            self.character_str = string.printable[:-6]
+            dict_character = list(self.character_str)
+        else:
+            self.character_str = None
+        assert self.character_str is not None, \
+            "Nonsupport type of the character: {}".format(self.character_str)
+        self.beg_str = "sos"
+        self.end_str = "eos"
+        if self.loss_type == "attention":
+            dict_character = [self.beg_str, self.end_str] + dict_character
+        self.dict = {}
+        for i, char in enumerate(dict_character):
+            self.dict[char] = i
+        self.character = dict_character
+
+    def encode(self, text):
+        """convert text-label into text-index.
+        input:
+            text: text labels of each image. [batch_size]
+
+        output:
+            text: concatenated text index for CTCLoss.
+                    [sum(text_lengths)] = [text_index_0 + text_index_1 + ... + text_index_(n - 1)]
+            length: length of each text. [batch_size]
+        """
+        if self.character_type == "en":
+            text = text.lower()
+
+        text_list = []
+        for char in text:
+            if char not in self.dict:
+                continue
+            text_list.append(self.dict[char])
+        text = np.array(text_list)
+        return text
+
+    def decode(self, text_index, is_remove_duplicate=False):
+        """ convert text-index into text-label. """
+        char_list = []
+        char_num = self.get_char_num()
+
+        if self.loss_type == "attention":
+            beg_idx = self.get_beg_end_flag_idx("beg")
+            end_idx = self.get_beg_end_flag_idx("end")
+            ignored_tokens = [beg_idx, end_idx]
+        else:
+            ignored_tokens = [char_num]
+
+        for idx in range(len(text_index)):
+            if text_index[idx] in ignored_tokens:
+                continue
+            if is_remove_duplicate:
+                if idx > 0 and text_index[idx - 1] == text_index[idx]:
+                    continue
+            char_list.append(self.character[text_index[idx]])
+        text = ''.join(char_list)
+        return text
+
+    def get_char_num(self):
+        return len(self.character)
+
+    def get_beg_end_flag_idx(self, beg_or_end):
+        if self.loss_type == "attention":
+            if beg_or_end == "beg":
+                idx = np.array(self.dict[self.beg_str])
+            elif beg_or_end == "end":
+                idx = np.array(self.dict[self.end_str])
+            else:
+                assert False, "Unsupport type %s in get_beg_end_flag_idx"\
+                    % beg_or_end
+            return idx
+        else:
+            err = "error in get_beg_end_flag_idx when using the loss %s"\
+                % (self.loss_type)
+            assert False, err
+
+
+class OCRReader(object):
+    def __init__(self,
+                 algorithm="CRNN",
+                 image_shape=[3, 32, 320],
+                 char_type="ch",
+                 batch_num=1,
+                 char_dict_path="./ppocr_keys_v1.txt"):
+        self.rec_image_shape = image_shape
+        self.character_type = char_type
+        self.rec_batch_num = batch_num
+        char_ops_params = {}
+        char_ops_params["character_type"] = char_type
+        char_ops_params["character_dict_path"] = char_dict_path
+        char_ops_params['loss_type'] = 'ctc'
+        self.char_ops = CharacterOps(char_ops_params)
+        self.label_ops = CTCLabelDecode(char_ops_params)
+
+    def resize_norm_img(self, img, max_wh_ratio):
+        imgC, imgH, imgW = self.rec_image_shape
+        if self.character_type == "ch":
+            imgW = int(32 * max_wh_ratio)
+        h = img.shape[0]
+        w = img.shape[1]
+        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_image = resized_image.transpose((2, 0, 1)) / 255
+        resized_image -= 0.5
+        resized_image /= 0.5
+        padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
+
+        padding_im[:, :, 0:resized_w] = resized_image
+        return padding_im
+
+    def preprocess(self, img_list):
+        img_num = len(img_list)
+        norm_img_batch = []
+        max_wh_ratio = 0
+        for ino in range(img_num):
+            h, w = img_list[ino].shape[0:2]
+            wh_ratio = w * 1.0 / h
+            max_wh_ratio = max(max_wh_ratio, wh_ratio)
+
+        for ino in range(img_num):
+            norm_img = self.resize_norm_img(img_list[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()
+
+        return norm_img_batch[0]
+
+    def postprocess_old(self, outputs, with_score=False):
+        rec_res = []
+        rec_idx_lod = outputs["ctc_greedy_decoder_0.tmp_0.lod"]
+        rec_idx_batch = outputs["ctc_greedy_decoder_0.tmp_0"]
+        if with_score:
+            predict_lod = outputs["softmax_0.tmp_0.lod"]
+        for rno in range(len(rec_idx_lod) - 1):
+            beg = rec_idx_lod[rno]
+            end = rec_idx_lod[rno + 1]
+            if isinstance(rec_idx_batch, list):
+                rec_idx_tmp = [x[0] for x in rec_idx_batch[beg:end]]
+            else:  #nd array
+                rec_idx_tmp = rec_idx_batch[beg:end, 0]
+            preds_text = self.char_ops.decode(rec_idx_tmp)
+            if with_score:
+                beg = predict_lod[rno]
+                end = predict_lod[rno + 1]
+                if isinstance(outputs["softmax_0.tmp_0"], list):
+                    outputs["softmax_0.tmp_0"] = np.array(outputs[
+                        "softmax_0.tmp_0"]).astype(np.float32)
+                probs = outputs["softmax_0.tmp_0"][beg:end, :]
+                ind = np.argmax(probs, axis=1)
+                blank = probs.shape[1]
+                valid_ind = np.where(ind != (blank - 1))[0]
+                score = np.mean(probs[valid_ind, ind[valid_ind]])
+                rec_res.append([preds_text, score])
+            else:
+                rec_res.append([preds_text])
+        return rec_res
+
+    def postprocess(self, outputs, with_score=False):
+        preds = outputs["save_infer_model/scale_0.tmp_1"]
+        try:
+            preds = preds.numpy()
+        except:
+            pass
+        preds_idx = preds.argmax(axis=2)
+        preds_prob = preds.max(axis=2)
+        text = self.label_ops.decode(
+            preds_idx, preds_prob, is_remove_duplicate=True)
+        return text

deploy/pdserving/pipeline_http_client.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 numpy as np
+import requests
+import json
+import base64
+import os
+
+
+def cv2_to_base64(image):
+    return base64.b64encode(image).decode('utf8')
+
+
+url = "http://127.0.0.1:9999/ocr/prediction"
+test_img_dir = "../doc/imgs/"
+for idx, img_file in enumerate(os.listdir(test_img_dir)):
+    with open(os.path.join(test_img_dir, img_file), 'rb') as file:
+        image_data1 = file.read()
+
+    image = cv2_to_base64(image_data1)
+
+    for i in range(1):
+        data = {"key": ["image"], "value": [image]}
+        r = requests.post(url=url, data=json.dumps(data))
+        print(r.json())
+
+test_img_dir = "../doc/imgs/"
+print("==> total number of test imgs: ", len(os.listdir(test_img_dir)))

deploy/pdserving/pipeline_rpc_client.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.
+try:
+    from paddle_serving_server_gpu.pipeline import PipelineClient
+except ImportError:
+    from paddle_serving_server.pipeline import PipelineClient
+import numpy as np
+import requests
+import json
+import cv2
+import base64
+import os
+
+client = PipelineClient()
+client.connect(['127.0.0.1:18090'])
+
+
+def cv2_to_base64(image):
+    return base64.b64encode(image).decode('utf8')
+
+
+test_img_dir = "imgs/"
+for img_file in os.listdir(test_img_dir):
+    with open(os.path.join(test_img_dir, img_file), 'rb') as file:
+        image_data = file.read()
+    image = cv2_to_base64(image_data)
+
+for i in range(1):
+    ret = client.predict(feed_dict={"image": image}, fetch=["res"])
+    print(ret)
+    #print(ret)

deploy/pdserving/web_service.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.
+try:
+    from paddle_serving_server_gpu.web_service import WebService, Op
+except ImportError:
+    from paddle_serving_server.web_service import WebService, Op
+
+import logging
+import numpy as np
+import cv2
+import base64
+# from paddle_serving_app.reader import OCRReader
+from ocr_reader import OCRReader, DetResizeForTest
+from paddle_serving_app.reader import Sequential, ResizeByFactor
+from paddle_serving_app.reader import Div, Normalize, Transpose
+from paddle_serving_app.reader import DBPostProcess, FilterBoxes, GetRotateCropImage, SortedBoxes
+
+_LOGGER = logging.getLogger()
+
+
+class DetOp(Op):
+    def init_op(self):
+        self.det_preprocess = Sequential([
+            DetResizeForTest(), Div(255),
+            Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]), Transpose(
+                (2, 0, 1))
+        ])
+        self.filter_func = FilterBoxes(10, 10)
+        self.post_func = DBPostProcess({
+            "thresh": 0.3,
+            "box_thresh": 0.5,
+            "max_candidates": 1000,
+            "unclip_ratio": 1.5,
+            "min_size": 3
+        })
+
+    def preprocess(self, input_dicts, data_id, log_id):
+        (_, input_dict), = input_dicts.items()
+        data = base64.b64decode(input_dict["image"].encode('utf8'))
+        data = np.fromstring(data, np.uint8)
+        # Note: class variables(self.var) can only be used in process op mode
+        im = cv2.imdecode(data, cv2.IMREAD_COLOR)
+        self.im = im
+        self.ori_h, self.ori_w, _ = im.shape
+
+        det_img = self.det_preprocess(self.im)
+        _, self.new_h, self.new_w = det_img.shape
+        print("det image shape", det_img.shape)
+        return {"x": det_img[np.newaxis, :].copy()}, False, None, ""
+
+    def postprocess(self, input_dicts, fetch_dict, log_id):
+        print("input_dicts: ", input_dicts)
+        det_out = fetch_dict["save_infer_model/scale_0.tmp_1"]
+        ratio_list = [
+            float(self.new_h) / self.ori_h, float(self.new_w) / self.ori_w
+        ]
+        dt_boxes_list = self.post_func(det_out, [ratio_list])
+        dt_boxes = self.filter_func(dt_boxes_list[0], [self.ori_h, self.ori_w])
+        out_dict = {"dt_boxes": dt_boxes, "image": self.im}
+
+        print("out dict", out_dict["dt_boxes"])
+        return out_dict, None, ""
+
+
+class RecOp(Op):
+    def init_op(self):
+        self.ocr_reader = OCRReader(
+            char_dict_path="../../ppocr/utils/ppocr_keys_v1.txt")
+
+        self.get_rotate_crop_image = GetRotateCropImage()
+        self.sorted_boxes = SortedBoxes()
+
+    def preprocess(self, input_dicts, data_id, log_id):
+        (_, input_dict), = input_dicts.items()
+        im = input_dict["image"]
+        dt_boxes = input_dict["dt_boxes"]
+        dt_boxes = self.sorted_boxes(dt_boxes)
+        feed_list = []
+        img_list = []
+        max_wh_ratio = 0
+        for i, dtbox in enumerate(dt_boxes):
+            boximg = self.get_rotate_crop_image(im, dt_boxes[i])
+            img_list.append(boximg)
+            h, w = boximg.shape[0:2]
+            wh_ratio = w * 1.0 / h
+            max_wh_ratio = max(max_wh_ratio, wh_ratio)
+        _, w, h = self.ocr_reader.resize_norm_img(img_list[0],
+                                                  max_wh_ratio).shape
+
+        imgs = np.zeros((len(img_list), 3, w, h)).astype('float32')
+        for id, img in enumerate(img_list):
+            norm_img = self.ocr_reader.resize_norm_img(img, max_wh_ratio)
+            imgs[id] = norm_img
+        print("rec image shape", imgs.shape)
+        feed = {"x": imgs.copy()}
+        return feed, False, None, ""
+
+    def postprocess(self, input_dicts, fetch_dict, log_id):
+        rec_res = self.ocr_reader.postprocess(fetch_dict, with_score=True)
+        res_lst = []
+        for res in rec_res:
+            res_lst.append(res[0])
+        res = {"res": str(res_lst)}
+        return res, None, ""
+
+
+class OcrService(WebService):
+    def get_pipeline_response(self, read_op):
+        det_op = DetOp(name="det", input_ops=[read_op])
+        rec_op = RecOp(name="rec", input_ops=[det_op])
+        return rec_op
+
+
+uci_service = OcrService(name="ocr")
+uci_service.prepare_pipeline_config("config.yml")
+uci_service.run_service()

deploy/slim/prune/README.md

+
+## 介绍
+
+复杂的模型有利于提高模型的性能，但也导致模型中存在一定冗余，模型裁剪通过移出网络模型中的子模型来减少这种冗余，达到减少模型计算复杂度，提高模型推理性能的目的。
+本教程将介绍如何使用飞桨模型压缩库PaddleSlim做PaddleOCR模型的压缩。
+[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim)集成了模型剪枝、量化（包括量化训练和离线量化）、蒸馏和神经网络搜索等多种业界常用且领先的模型压缩功能，如果您感兴趣，可以关注并了解。
+
+
+在开始本教程之前，建议先了解：
+1. [PaddleOCR模型的训练方法](../../../doc/doc_ch/quickstart.md)
+2. [模型裁剪教程](https://github.com/PaddlePaddle/PaddleSlim/blob/release%2F2.0.0/docs/zh_cn/tutorials/pruning/dygraph/filter_pruning.md)
+
+
+## 快速开始
+
+模型裁剪主要包括四个步骤：
+1. 安装 PaddleSlim
+2. 准备训练好的模型
+3. 敏感度分析、裁剪训练
+4. 导出模型、预测部署
+
+### 1. 安装PaddleSlim
+
+```bash
+git clone https://github.com/PaddlePaddle/PaddleSlim.git
+git checkout develop
+cd Paddleslim
+python3 setup.py install
+```
+
+### 2. 获取预训练模型
+模型裁剪需要加载事先训练好的模型，PaddleOCR也提供了一系列(模型)[../../../doc/doc_ch/models_list.md]，开发者可根据需要自行选择模型或使用自己的模型。
+
+### 3. 敏感度分析训练
+
+加载预训练模型后，通过对现有模型的每个网络层进行敏感度分析，得到敏感度文件：sen.pickle，可以通过PaddleSlim提供的[接口](https://github.com/PaddlePaddle/PaddleSlim/blob/9b01b195f0c4bc34a1ab434751cb260e13d64d9e/paddleslim/dygraph/prune/filter_pruner.py#L75)加载文件，获得各网络层在不同裁剪比例下的精度损失。从而了解各网络层冗余度，决定每个网络层的裁剪比例。
+敏感度文件内容格式：
+    sen.pickle(Dict){
+            'layer_weight_name_0': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+            'layer_weight_name_1': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+        }
+
+    例子：
+        {
+            'conv10_expand_weights': {0.1: 0.006509952684312718, 0.2: 0.01827734339798862, 0.3: 0.014528405644659832, 0.6: 0.06536008804270439, 0.8: 0.11798612250664964, 0.7: 0.12391408417493704, 0.4: 0.030615754498018757, 0.5: 0.047105205602406594}
+            'conv10_linear_weights': {0.1: 0.05113190831455035, 0.2: 0.07705573833558801, 0.3: 0.12096721757739311, 0.6: 0.5135061352930738, 0.8: 0.7908166677143281, 0.7: 0.7272187676899062, 0.4: 0.1819252083008504, 0.5: 0.3728054727792405}
+        }
+加载敏感度文件后会返回一个字典，字典中的keys为网络模型参数模型的名字，values为一个字典，里面保存了相应网络层的裁剪敏感度信息。例如在例子中，conv10_expand_weights所对应的网络层在裁掉10%的卷积核后模型性能相较原模型会下降0.65%，详细信息可见[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/zh_cn/algo/algo.md#2-%E5%8D%B7%E7%A7%AF%E6%A0%B8%E5%89%AA%E8%A3%81%E5%8E%9F%E7%90%86)
+
+进入PaddleOCR根目录，通过以下命令对模型进行敏感度分析训练：
+```bash
+python3.7 deploy/slim/prune/sensitivity_anal.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrain_weights="your trained model"
+```
+
+### 4. 导出模型、预测部署
+
+在得到裁剪训练保存的模型后，我们可以将其导出为inference_model：
+```bash
+pytho3.7 deploy/slim/prune/export_prune_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrain_weights=./output/det_db/best_accuracy  Global.save_inference_dir=inference_model
+```
+
+inference model的预测和部署参考：
+1. [inference model python端预测](../../../doc/doc_ch/inference.md)
+2. [inference model C++预测](../../cpp_infer/readme.md)

deploy/slim/prune/README_en.md

+
+## Introduction
+
+Generally, a more complex model would achive better performance in the task, but it also leads to some redundancy in the model. Model Pruning is a technique that reduces this redundancy by removing the sub-models in the neural network model, so as to reduce model calculation complexity and improve model inference performance.
+
+This example uses PaddleSlim provided[APIs of Pruning](https://paddlepaddle.github.io/PaddleSlim/api/prune_api/) to compress the OCR model.
+[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim), an open source library which integrates model pruning, quantization (including quantization training and offline quantization), distillation, neural network architecture search, and many other commonly used and leading model compression technique in the industry.
+
+It is recommended that you could understand following pages before reading this example：
+1. [PaddleOCR training methods](../../../doc/doc_ch/quickstart.md)
+2. [The demo of prune](https://github.com/PaddlePaddle/PaddleSlim/blob/release%2F2.0.0/docs/zh_cn/tutorials/pruning/dygraph/filter_pruning.md)
+
+## Quick start
+
+Five steps for OCR model prune:
+1. Install PaddleSlim
+2. Prepare the trained model
+3. Sensitivity analysis and tailoring training
+4. Export model, predict deployment
+
+### 1. Install PaddleSlim
+
+```bash
+git clone https://github.com/PaddlePaddle/PaddleSlim.git
+git checkout develop
+cd Paddleslim
+python3 setup.py install
+```
+
+
+### 2. Download Pretrain Model
+Model prune needs to load pre-trained models.
+PaddleOCR also provides a series of (models)[../../../doc/doc_en/models_list_en.md]. Developers can choose their own models or use their own models according to their needs.
+
+
+### 3. Pruning sensitivity analysis
+
+  After the pre-training model is loaded, sensitivity analysis is performed on each network layer of the model to understand the redundancy of each network layer, and save a sensitivity file which named: sen.pickle.  After that, user could load the sensitivity file via the [methods provided by PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/paddleslim/prune/sensitive.py#L221) and determining the pruning ratio of each network layer automatically. For specific details of sensitivity analysis, see：[Sensitivity analysis](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/zh_cn/tutorials/image_classification_sensitivity_analysis_tutorial.md)
+  The data format of sensitivity file：
+      sen.pickle(Dict){
+              'layer_weight_name_0': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+              'layer_weight_name_1': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+          }
+
+      example：
+          {
+              'conv10_expand_weights': {0.1: 0.006509952684312718, 0.2: 0.01827734339798862, 0.3: 0.014528405644659832, 0.6: 0.06536008804270439, 0.8: 0.11798612250664964, 0.7: 0.12391408417493704, 0.4: 0.030615754498018757, 0.5: 0.047105205602406594}
+              'conv10_linear_weights': {0.1: 0.05113190831455035, 0.2: 0.07705573833558801, 0.3: 0.12096721757739311, 0.6: 0.5135061352930738, 0.8: 0.7908166677143281, 0.7: 0.7272187676899062, 0.4: 0.1819252083008504, 0.5: 0.3728054727792405}
+          }
+  The function would return a dict after loading the sensitivity file. The keys of the dict are name of parameters in each layer. And the value of key is the information about pruning sensitivity of correspoding layer. In example, pruning 10% filter of the layer corresponding to conv10_expand_weights would lead to 0.65% degradation of model performance. The details could be seen at: [Sensitivity analysis](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/zh_cn/algo/algo.md#2-%E5%8D%B7%E7%A7%AF%E6%A0%B8%E5%89%AA%E8%A3%81%E5%8E%9F%E7%90%86)
+
+
+Enter the PaddleOCR root directory，perform sensitivity analysis on the model with the following command：
+
+```bash
+
+python3.7 deploy/slim/prune/sensitivity_anal.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrain_weights="your trained model"
+
+```
+
+
+### 5.  Export inference model and deploy it
+
+We can export the pruned model as inference_model for deployment:
+```bash
+python deploy/slim/prune/export_prune_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml  -o Global.pretrain_weights=./output/det_db/best_accuracy Global.test_batch_size_per_card=1 Global.save_inference_dir=inference_model
+```
+
+Reference for prediction and deployment of inference model:
+1. [inference model python prediction](../../../doc/doc_en/inference_en.md)
+2. [inference model C++ prediction](../../cpp_infer/readme_en.md)

deploy/slim/prune/export_prune_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
+
+import os
+import sys
+
+__dir__ = os.path.dirname(__file__)
+sys.path.append(__dir__)
+sys.path.append(os.path.join(__dir__, '..', '..', '..'))
+sys.path.append(os.path.join(__dir__, '..', '..', '..', 'tools'))
+
+import paddle
+from ppocr.data import build_dataloader
+from ppocr.modeling.architectures import build_model
+
+from ppocr.postprocess import build_post_process
+from ppocr.metrics import build_metric
+from ppocr.utils.save_load import init_model
+import tools.program as program
+
+
+def main(config, device, logger, vdl_writer):
+
+    global_config = config['Global']
+
+    # build dataloader
+    valid_dataloader = build_dataloader(config, 'Eval', device, logger)
+
+    # build post process
+    post_process_class = build_post_process(config['PostProcess'],
+                                            global_config)
+
+    # build model
+    # for rec algorithm
+    if hasattr(post_process_class, 'character'):
+        char_num = len(getattr(post_process_class, 'character'))
+        config['Architecture']["Head"]['out_channels'] = char_num
+    model = build_model(config['Architecture'])
+
+    flops = paddle.flops(model, [1, 3, 640, 640])
+    logger.info(f"FLOPs before pruning: {flops}")
+
+    from paddleslim.dygraph import FPGMFilterPruner
+    model.train()
+    pruner = FPGMFilterPruner(model, [1, 3, 640, 640])
+
+    # build metric
+    eval_class = build_metric(config['Metric'])
+
+    def eval_fn():
+        metric = program.eval(model, valid_dataloader, post_process_class,
+                              eval_class)
+        logger.info(f"metric['hmean']: {metric['hmean']}")
+        return metric['hmean']
+
+    params_sensitive = pruner.sensitive(
+        eval_func=eval_fn,
+        sen_file="./sen.pickle",
+        skip_vars=[
+            "conv2d_57.w_0", "conv2d_transpose_2.w_0", "conv2d_transpose_3.w_0"
+        ])
+
+    logger.info(
+        "The sensitivity analysis results of model parameters saved in sen.pickle"
+    )
+    # calculate pruned params's ratio
+    params_sensitive = pruner._get_ratios_by_loss(params_sensitive, loss=0.02)
+    for key in params_sensitive.keys():
+        logger.info(f"{key}, {params_sensitive[key]}")
+
+    plan = pruner.prune_vars(params_sensitive, [0])
+
+    flops = paddle.flops(model, [1, 3, 640, 640])
+    logger.info(f"FLOPs after pruning: {flops}")
+
+    # load pretrain model
+    pre_best_model_dict = init_model(config, model, logger, None)
+    metric = program.eval(model, valid_dataloader, post_process_class,
+                          eval_class)
+    logger.info(f"metric['hmean']: {metric['hmean']}")
+
+    # start export model
+    from paddle.jit import to_static
+
+    infer_shape = [3, -1, -1]
+    if config['Architecture']['model_type'] == "rec":
+        infer_shape = [3, 32, -1]  # for rec model, H must be 32
+
+        if 'Transform' in config['Architecture'] and config['Architecture'][
+                'Transform'] is not None and config['Architecture'][
+                    'Transform']['name'] == 'TPS':
+            logger.info(
+                'When there is tps in the network, variable length input is not supported, and the input size needs to be the same as during training'
+            )
+            infer_shape[-1] = 100
+    model = to_static(
+        model,
+        input_spec=[
+            paddle.static.InputSpec(
+                shape=[None] + infer_shape, dtype='float32')
+        ])
+
+    save_path = '{}/inference'.format(config['Global']['save_inference_dir'])
+    paddle.jit.save(model, save_path)
+    logger.info('inference model is saved to {}'.format(save_path))
+
+
+if __name__ == '__main__':
+    config, device, logger, vdl_writer = program.preprocess(is_train=True)
+    main(config, device, logger, vdl_writer)

deploy/slim/prune/sensitivity_anal.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
+
+import os
+import sys
+
+__dir__ = os.path.dirname(__file__)
+sys.path.append(__dir__)
+sys.path.append(os.path.join(__dir__, '..', '..', '..'))
+sys.path.append(os.path.join(__dir__, '..', '..', '..', 'tools'))
+
+import paddle
+import paddle.distributed as dist
+from ppocr.data import build_dataloader
+from ppocr.modeling.architectures import build_model
+from ppocr.losses import build_loss
+from ppocr.optimizer import build_optimizer
+from ppocr.postprocess import build_post_process
+from ppocr.metrics import build_metric
+from ppocr.utils.save_load import init_model
+import tools.program as program
+
+dist.get_world_size()
+
+
+def get_pruned_params(parameters):
+    params = []
+
+    for param in parameters:
+        if len(
+                param.shape
+        ) == 4 and 'depthwise' not in param.name and 'transpose' not in param.name and "conv2d_57" not in param.name and "conv2d_56" not in param.name:
+            params.append(param.name)
+    return params
+
+
+def main(config, device, logger, vdl_writer):
+    # init dist environment
+    if config['Global']['distributed']:
+        dist.init_parallel_env()
+
+    global_config = config['Global']
+
+    # build dataloader
+    train_dataloader = build_dataloader(config, 'Train', device, logger)
+    if config['Eval']:
+        valid_dataloader = build_dataloader(config, 'Eval', device, logger)
+    else:
+        valid_dataloader = None
+
+    # build post process
+    post_process_class = build_post_process(config['PostProcess'],
+                                            global_config)
+
+    # build model
+    # for rec algorithm
+    if hasattr(post_process_class, 'character'):
+        char_num = len(getattr(post_process_class, 'character'))
+        config['Architecture']["Head"]['out_channels'] = char_num
+    model = build_model(config['Architecture'])
+
+    flops = paddle.flops(model, [1, 3, 640, 640])
+    logger.info(f"FLOPs before pruning: {flops}")
+
+    from paddleslim.dygraph import FPGMFilterPruner
+    model.train()
+    pruner = FPGMFilterPruner(model, [1, 3, 640, 640])
+
+    # build loss
+    loss_class = build_loss(config['Loss'])
+
+    # build optim
+    optimizer, lr_scheduler = build_optimizer(
+        config['Optimizer'],
+        epochs=config['Global']['epoch_num'],
+        step_each_epoch=len(train_dataloader),
+        parameters=model.parameters())
+
+    # build metric
+    eval_class = build_metric(config['Metric'])
+    # load pretrain model
+    pre_best_model_dict = init_model(config, model, logger, optimizer)
+
+    logger.info('train dataloader has {} iters, valid dataloader has {} iters'.
+                format(len(train_dataloader), len(valid_dataloader)))
+    # build metric
+    eval_class = build_metric(config['Metric'])
+
+    logger.info('train dataloader has {} iters, valid dataloader has {} iters'.
+                format(len(train_dataloader), len(valid_dataloader)))
+
+    def eval_fn():
+        metric = program.eval(model, valid_dataloader, post_process_class,
+                              eval_class)
+        logger.info(f"metric['hmean']: {metric['hmean']}")
+        return metric['hmean']
+
+    params_sensitive = pruner.sensitive(
+        eval_func=eval_fn,
+        sen_file="./sen.pickle",
+        skip_vars=[
+            "conv2d_57.w_0", "conv2d_transpose_2.w_0", "conv2d_transpose_3.w_0"
+        ])
+
+    logger.info(
+        "The sensitivity analysis results of model parameters saved in sen.pickle"
+    )
+    # calculate pruned params's ratio
+    params_sensitive = pruner._get_ratios_by_loss(params_sensitive, loss=0.02)
+    for key in params_sensitive.keys():
+        logger.info(f"{key}, {params_sensitive[key]}")
+
+    plan = pruner.prune_vars(params_sensitive, [0])
+    for param in model.parameters():
+        if ("weights" in param.name and "conv" in param.name) or (
+                "w_0" in param.name and "conv2d" in param.name):
+            logger.info(f"{param.name}: {param.shape}")
+
+    flops = paddle.flops(model, [1, 3, 640, 640])
+    logger.info(f"FLOPs after pruning: {flops}")
+
+    # start train
+
+    program.train(config, train_dataloader, valid_dataloader, device, model,
+                  loss_class, optimizer, lr_scheduler, post_process_class,
+                  eval_class, pre_best_model_dict, logger, vdl_writer)
+
+
+if __name__ == '__main__':
+    config, device, logger, vdl_writer = program.preprocess(is_train=True)
+    main(config, device, logger, vdl_writer)

doc/doc_ch/algorithm_overview.md

@@ -28,7 +28,9 @@ PaddleOCR开源的文本检测算法列表：
 | --- | --- | --- | --- | --- | --- |
 |SAST|ResNet50_vd|89.63%|78.44%|83.66%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_totaltext_v2.0_train.tar)|
 
-**说明：** SAST模型训练额外加入了icdar2013、icdar2017、COCO-Text、ArT等公开数据集进行调优。PaddleOCR用到的经过整理格式的英文公开数据集下载：[百度云地址](https://pan.baidu.com/s/12cPnZcVuV1zn5DOd4mqjVw) (提取码: 2bpi)
+**说明：** SAST模型训练额外加入了icdar2013、icdar2017、COCO-Text、ArT等公开数据集进行调优。PaddleOCR用到的经过整理格式的英文公开数据集下载：
+* [百度云地址](https://pan.baidu.com/s/12cPnZcVuV1zn5DOd4mqjVw) (提取码: 2bpi)
+* [Google Drive下载地址](https://drive.google.com/drive/folders/1ll2-XEVyCQLpJjawLDiRlvo_i4BqHCJe?usp=sharing)
 
 PaddleOCR文本检测算法的训练和使用请参考文档教程中[模型训练/评估中的文本检测部分](./detection.md)。
 

doc/doc_ch/angle_class.md

 ## 文字角度分类
+### 方法介绍
+文字角度分类主要用于图片非0度的场景下，在这种场景下需要对图片里检测到的文本行进行一个转正的操作。在PaddleOCR系统内，
+文字检测之后得到的文本行图片经过仿射变换之后送入识别模型，此时只需要对文字进行一个0和180度的角度分类，因此PaddleOCR内置的
+文字角度分类器**只支持了0和180度的分类**。如果想支持更多角度，可以自己修改算法进行支持。
+
+0和180度数据样本例子：
+
+![](../imgs_results/angle_class_example.jpg)
 
 ### 数据准备
 
@@ -13,7 +21,7 @@ ln -sf <path/to/dataset> <path/to/paddle_ocr>/train_data/cls/dataset
 请参考下文组织您的数据。
 - 训练集
 
-首先请将训练图片放入同一个文件夹（train_images），并用一个txt文件（cls_gt_train.txt）记录图片路径和标签。
+首先建议将训练图片放入同一个文件夹，并用一个txt文件（cls_gt_train.txt）记录图片路径和标签。
 
 **注意：** 默认请将图片路径和图片标签用 `\t` 分割，如用其他方式分割将造成训练报错
 
@@ -21,8 +29,8 @@ ln -sf <path/to/dataset> <path/to/paddle_ocr>/train_data/cls/dataset
 
 ```
 " 图像文件名                 图像标注信息 "
-train/word_001.jpg   0
-train/word_002.jpg   180
+train/cls/train/word_001.jpg   0
+train/cls/train/word_002.jpg   180
 ```
 
 最终训练集应有如下文件结构：

doc/doc_ch/inference.md

@@ -2,16 +2,17 @@
 # 基于Python预测引擎推理
 
 inference 模型（`paddle.jit.save`保存的模型）
-一般是模型训练完成后保存的固化模型，多用于预测部署。训练过程中保存的模型是checkpoints模型，保存的是模型的参数，多用于恢复训练等。
-与checkpoints模型相比，inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合与实际系统集成。
+一般是模型训练，把模型结构和模型参数保存在文件中的固化模型，多用于预测部署场景。
+训练过程中保存的模型是checkpoints模型，保存的只有模型的参数，多用于恢复训练等。
+与checkpoints模型相比，inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合于实际系统集成。
 
-接下来首先介绍如何将训练的模型转换成inference模型，然后将依次介绍文本检测、文本角度分类器、文本识别以及三者串联基于预测引擎推理。
+接下来首先介绍如何将训练的模型转换成inference模型，然后将依次介绍文本检测、文本角度分类器、文本识别以及三者串联在CPU、GPU上的预测方法。
 
 
 - [一、训练模型转inference模型](#训练模型转inference模型)
     - [检测模型转inference模型](#检测模型转inference模型)
     - [识别模型转inference模型](#识别模型转inference模型)  
-    - [方向分类模型转inference模型](#方向分类模型转inference模型)  
+    - [方向分类模型转inference模型](#方向分类模型转inference模型)
 
 - [二、文本检测模型推理](#文本检测模型推理)
     - [1. 超轻量中文检测模型推理](#超轻量中文检测模型推理)
@@ -140,7 +141,7 @@ python3 tools/infer/predict_det.py --image_dir="./doc/imgs/00018069.jpg" --det_m
 ![](../imgs_results/det_res_00018069.jpg)
 
 通过参数`limit_type`和`det_limit_side_len`来对图片的尺寸进行限制，
-`litmit_type`可选参数为[`max`, `min`]，
+`limit_type`可选参数为[`max`, `min`]，
 `det_limit_size_len` 为正整数，一般设置为32 的倍数，比如960。
 
 参数默认设置为`limit_type='max', det_limit_side_len=960`。表示网络输入图像的最长边不能超过960，

doc/doc_ch/installation.md

@@ -30,7 +30,7 @@ sudo nvidia-docker run --name ppocr -v $PWD:/paddle --shm-size=64G --network=hos
 sudo docker container exec -it ppocr /bin/bash
 ```
 
-**2. 安装PaddlePaddle Fluid v2.0**
+**2. 安装PaddlePaddle 2.0**
 ```
 pip3 install --upgrade pip
 

doc/doc_ch/models_list.md

@@ -104,27 +104,16 @@ python3 generate_multi_language_configs.py -l it \
 | german_mobile_v2.0_rec |德文识别|[rec_german_lite_train.yml](../../configs/rec/multi_language/rec_german_lite_train.yml)|2.65M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/german_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/german_mobile_v2.0_rec_train.tar) |
 | korean_mobile_v2.0_rec |韩文识别|[rec_korean_lite_train.yml](../../configs/rec/multi_language/rec_korean_lite_train.yml)|3.9M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/korean_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/korean_mobile_v2.0_rec_train.tar) |
 | japan_mobile_v2.0_rec |日文识别|[rec_japan_lite_train.yml](../../configs/rec/multi_language/rec_japan_lite_train.yml)|4.23M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/japan_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/japan_mobile_v2.0_rec_train.tar) |
-| it_mobile_v2.0_rec |意大利文识别|rec_it_lite_train.yml|2.53M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/it_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/it_mobile_v2.0_rec_train.tar) |
-| xi_mobile_v2.0_rec |西班牙文识别|rec_xi_lite_train.yml|2.53M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/xi_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/xi_mobile_v2.0_rec_train.tar) |
-| pu_mobile_v2.0_rec |葡萄牙文识别|rec_pu_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/pu_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/pu_mobile_v2.0_rec_train.tar) |
-| ru_mobile_v2.0_rec |俄罗斯文识别|rec_ru_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ru_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ru_mobile_v2.0_rec_train.tar) |
-| ar_mobile_v2.0_rec |阿拉伯文识别|rec_ar_lite_train.yml|2.53M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ar_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ar_mobile_v2.0_rec_train.tar) |
-| hi_mobile_v2.0_rec |印地文识别|rec_hi_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/hi_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/hi_mobile_v2.0_rec_train.tar) |
 | chinese_cht_mobile_v2.0_rec |中文繁体识别|rec_chinese_cht_lite_train.yml|5.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/chinese_cht_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/chinese_cht_mobile_v2.0_rec_train.tar) |
-| ug_mobile_v2.0_rec |维吾尔文识别|rec_ug_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ug_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ug_mobile_v2.0_rec_train.tar) |
-| fa_mobile_v2.0_rec |波斯文识别|rec_fa_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/fa_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/fa_mobile_v2.0_rec_train.tar) |
-| ur_mobile_v2.0_rec |乌尔都文识别|rec_ur_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ur_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ur_mobile_v2.0_rec_train.tar) |
-| rs_mobile_v2.0_rec |塞尔维亚文（latin）识别|rec_rs_lite_train.yml|2.53M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/rs_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/rs_mobile_v2.0_rec_train.tar) |
-| oc_mobile_v2.0_rec |欧西坦文识别|rec_oc_lite_train.yml|2.53M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/oc_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/oc_mobile_v2.0_rec_train.tar) |
-| mr_mobile_v2.0_rec |马拉地文识别|rec_mr_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/mr_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/mr_mobile_v2.0_rec_train.tar) |
-| ne_mobile_v2.0_rec |尼泊尔文识别|rec_ne_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ne_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ne_mobile_v2.0_rec_train.tar) |
-| rsc_mobile_v2.0_rec |塞尔维亚文（cyrillic）识别|rec_rsc_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/rsc_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/rsc_mobile_v2.0_rec_train.tar) |
-| bg_mobile_v2.0_rec |保加利亚文识别|rec_bg_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/bg_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/bg_mobile_v2.0_rec_train.tar) |
-| uk_mobile_v2.0_rec |乌克兰文识别|rec_uk_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/uk_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/uk_mobile_v2.0_rec_train.tar) |
-| be_mobile_v2.0_rec |白俄罗斯文识别|rec_be_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/be_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/be_mobile_v2.0_rec_train.tar) |
 | te_mobile_v2.0_rec |泰卢固文识别|rec_te_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/te_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/te_mobile_v2.0_rec_train.tar) |
 | ka_mobile_v2.0_rec |卡纳达文识别|rec_ka_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ka_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ka_mobile_v2.0_rec_train.tar) |
 | ta_mobile_v2.0_rec |泰米尔文识别|rec_ta_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ta_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ta_mobile_v2.0_rec_train.tar) |
+| latin_mobile_v2.0_rec |  拉丁文识别 |  [rec_latin_lite_train.yml](../../configs/rec/multi_language/rec_latin_lite_train.yml) |2.6M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/latin_ppocr_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/latin_ppocr_mobile_v2.0_rec_train.tar) |
+| arabic_mobile_v2.0_rec |  阿拉伯字母 | [rec_arabic_lite_train.yml](../../configs/rec/multi_language/rec_arabic_lite_train.yml) |2.6M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/arabic_ppocr_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/arabic_ppocr_mobile_v2.0_rec_train.tar) |
+| cyrillic_mobile_v2.0_rec | 斯拉夫字母 | [rec_cyrillic_lite_train.yml](../../configs/rec/multi_language/rec_cyrillic_lite_train.yml) |2.6M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/cyrillic_ppocr_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/cyrillic_ppocr_mobile_v2.0_rec_train.tar) |
+| devanagari_mobile_v2.0_rec | 梵文字母 | [rec_devanagari_lite_train.yml](../../configs/rec/multi_language/rec_devanagari_lite_train.yml) |2.6M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/devanagari_ppocr_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/devanagari_ppocr_mobile_v2.0_rec_train.tar) |
+
+更多支持语种请参考: [多语言模型](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.1/doc/doc_ch/multi_languages.md#%E8%AF%AD%E7%A7%8D%E7%BC%A9%E5%86%99)
 
 
 <a name="文本方向分类模型"></a>

doc/doc_ch/multi_languages.md

+# 多语言模型
+
+**近期更新**
+
+- 2021.4.9 支持**80种**语言的检测和识别
+- 2021.4.9 支持**轻量高精度**英文模型检测识别
+
+PaddleOCR 旨在打造一套丰富、领先、且实用的OCR工具库，不仅提供了通用场景下的中英文模型，也提供了专门在英文场景下训练的模型，
+和覆盖[80个语言](#语种缩写)的小语种模型。
+
+其中英文模型支持，大小写字母和常见标点的检测识别，并优化了空格字符的识别：
+
+<div align="center">
+    <img src="../imgs_results/multi_lang/en_1.jpg" width="400" height="600">
+</div>
+
+小语种模型覆盖了拉丁语系、阿拉伯语系、中文繁体、韩语、日语等等：
+
+<div align="center">
+    <img src="../imgs_results/multi_lang/japan_2.jpg" width="600" height="300">
+    <img src="../imgs_results/multi_lang/french_0.jpg" width="300" height="300">
+</div>
+
+
+本文档将简要介绍小语种模型的使用方法。
+
+- [1 安装](#安装)
+    - [1.1 paddle 安装](#paddle安装)
+    - [1.2 paddleocr package 安装](#paddleocr_package_安装)  
+
+- [2 快速使用](#快速使用)
+    - [2.1 命令行运行](#命令行运行)
+     - [2.1.1 整图预测](#bash_检测+识别)
+     - [2.1.2 识别预测](#bash_识别)
+     - [2.1.3 检测预测](#bash_检测)
+    - [2.2 python 脚本运行](#python_脚本运行)
+     - [2.2.1 整图预测](#python_检测+识别)
+     - [2.2.2 识别预测](#python_识别)
+     - [2.2.3 检测预测](#python_检测)
+- [3 自定义训练](#自定义训练)
+- [4 支持语种及缩写](#语种缩写)
+
+<a name="安装"></a>
+## 1 安装
+
+<a name="paddle安装"></a>
+### 1.1 paddle 安装
+```
+# cpu
+pip install paddlepaddle
+
+# gpu
+pip instll paddlepaddle-gpu
+```
+
+<a name="paddleocr_package_安装"></a>
+### 1.2 paddleocr package 安装
+
+
+pip 安装
+```
+pip install "paddleocr>=2.0.6" # 推荐使用2.0.6版本
+```
+本地构建并安装
+```
+python3 setup.py bdist_wheel
+pip3 install dist/paddleocr-x.x.x-py3-none-any.whl # x.x.x是paddleocr的版本号
+```
+
+<a name="快速使用"></a>
+## 2 快速使用
+
+<a name="命令行运行"></a>
+### 2.1 命令行运行
+
+查看帮助信息
+
+```
+paddleocr -h
+```
+
+* 整图预测（检测+识别）
+
+Paddleocr目前支持80个语种，可以通过修改--lang参数进行切换，具体支持的[语种](#语种缩写)可查看表格。
+
+``` bash
+
+paddleocr --image_dir doc/imgs/japan_2.jpg --lang=japan
+```
+
+<div align="center">
+    <img src="https://raw.githubusercontent.com/PaddlePaddle/PaddleOCR/release/2.1/doc/imgs/japan_2.jpg" width="800">
+</div>
+
+
+结果是一个list，每个item包含了文本框，文字和识别置信度
+```text
+[[[671.0, 60.0], [847.0, 63.0], [847.0, 104.0], [671.0, 102.0]], ('もちもち', 0.9993342)]
+[[[394.0, 82.0], [536.0, 77.0], [538.0, 127.0], [396.0, 132.0]], ('天然の', 0.9919842)]
+[[[880.0, 89.0], [1014.0, 93.0], [1013.0, 127.0], [879.0, 124.0]], ('とろっと', 0.9976762)]
+[[[1067.0, 101.0], [1294.0, 101.0], [1294.0, 138.0], [1067.0, 138.0]], ('後味のよい', 0.9988712)]
+......
+```
+
+* 识别预测
+
+```bash
+paddleocr --image_dir doc/imgs_words/japan/1.jpg   --det false --lang=japan
+```
+
+![](https://raw.githubusercontent.com/PaddlePaddle/PaddleOCR/release/2.0/doc/imgs_words/japan/1.jpg)
+
+结果是一个tuple，返回识别结果和识别置信度
+
+```text
+('したがって', 0.99965394)
+```
+
+* 检测预测
+
+```
+paddleocr --image_dir PaddleOCR/doc/imgs/11.jpg --rec false
+```
+
+结果是一个list，每个item只包含文本框
+
+```
+[[26.0, 457.0], [137.0, 457.0], [137.0, 477.0], [26.0, 477.0]]
+[[25.0, 425.0], [372.0, 425.0], [372.0, 448.0], [25.0, 448.0]]
+[[128.0, 397.0], [273.0, 397.0], [273.0, 414.0], [128.0, 414.0]]
+......
+```
+
+<a name="python_脚本运行"></a>
+### 2.2 python 脚本运行
+
+ppocr 也支持在python脚本中运行，便于嵌入到您自己的代码中：
+
+* 整图预测（检测+识别）
+
+```
+from paddleocr import PaddleOCR, draw_ocr
+
+# 同样也是通过修改 lang 参数切换语种
+ocr = PaddleOCR(lang="korean") # 首次执行会自动下载模型文件
+img_path = 'doc/imgs/korean_1.jpg '
+result = ocr.ocr(img_path)
+# 打印检测框和识别结果
+for line in result:
+    print(line)
+
+# 可视化
+from PIL import Image
+image = Image.open(img_path).convert('RGB')
+boxes = [line[0] for line in result]
+txts = [line[1][0] for line in result]
+scores = [line[1][1] for line in result]
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/korean.ttf')
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+结果可视化:
+
+<div align="center">
+    <img src="https://raw.githubusercontent.com/PaddlePaddle/PaddleOCR/release/2.1/doc/imgs_results/korean.jpg" width="800">
+</div>
+
+* 识别预测
+
+```
+from paddleocr import PaddleOCR
+ocr = PaddleOCR(lang="german")
+img_path = 'PaddleOCR/doc/imgs_words/german/1.jpg'
+result = ocr.ocr(img_path, det=False, cls=True)
+for line in result:
+    print(line)
+```
+
+
+![](../imgs_words/german/1.jpg)
+
+结果是一个tuple，只包含识别结果和识别置信度
+
+```
+('leider auch jetzt', 0.97538936)
+```
+
+* 检测预测
+
+```python
+from paddleocr import PaddleOCR, draw_ocr
+ocr = PaddleOCR() # need to run only once to download and load model into memory
+img_path = 'PaddleOCR/doc/imgs_en/img_12.jpg'
+result = ocr.ocr(img_path, rec=False)
+for line in result:
+    print(line)
+
+# 显示结果
+from PIL import Image
+
+image = Image.open(img_path).convert('RGB')
+im_show = draw_ocr(image, result, txts=None, scores=None, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+结果是一个list，每个item只包含文本框
+```bash
+[[26.0, 457.0], [137.0, 457.0], [137.0, 477.0], [26.0, 477.0]]
+[[25.0, 425.0], [372.0, 425.0], [372.0, 448.0], [25.0, 448.0]]
+[[128.0, 397.0], [273.0, 397.0], [273.0, 414.0], [128.0, 414.0]]
+......
+```
+
+结果可视化 ：
+
+<div align="center">
+    <img src="https://raw.githubusercontent.com/PaddlePaddle/PaddleOCR/release/2.1/doc/imgs_results/whl/12_det.jpg" width="800">
+</div>
+
+ppocr 还支持方向分类， 更多使用方式请参考：[whl包使用说明](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.0/doc/doc_ch/whl.md)。
+
+<a name="自定义训练"></a>
+## 3 自定义训练
+
+ppocr 支持使用自己的数据进行自定义训练或finetune, 其中识别模型可以参考 [法语配置文件](../../configs/rec/multi_language/rec_french_lite_train.yml)
+修改训练数据路径、字典等参数。
+
+具体数据准备、训练过程可参考：[文本检测](../doc_ch/detection.md)、[文本识别](../doc_ch/recognition.md)，更多功能如预测部署、
+数据标注等功能可以阅读完整的[文档教程](../../README_ch.md)。
+
+<a name="语种缩写"></a>
+## 4 支持语种及缩写
+
+| 语种 | 描述 | 缩写 |
+| --- | --- | --- |
+|中文|chinese and english|ch|
+|英文|english|en|
+|法文|french|fr|
+|德文|german|german|
+|日文|japan|japan|
+|韩文|korean|korean|
+|中文繁体|chinese traditional |ch_tra|
+|意大利文| Italian |it|
+|西班牙文|Spanish |es|
+|葡萄牙文| Portuguese|pt|
+|俄罗斯文|Russia|ru|
+|阿拉伯文|Arabic|ar|
+|印地文|Hindi|hi|
+|维吾尔|Uyghur|ug|
+|波斯文|Persian|fa|
+|乌尔都文|Urdu|ur|
+|塞尔维亚文（latin)| Serbian(latin) |rs_latin|
+|欧西坦文|Occitan |oc|
+|马拉地文|Marathi|mr|
+|尼泊尔文|Nepali|ne|
+|塞尔维亚文（cyrillic)|Serbian(cyrillic)|rs_cyrillic|
+|保加利亚文|Bulgarian |bg|
+|乌克兰文|Ukranian|uk|
+|白俄罗斯文|Belarusian|be|
+|泰卢固文|Telugu |te|
+|卡纳达文|Kannada |kn|
+|泰米尔文|Tamil |ta|
+|南非荷兰文 |Afrikaans |af|
+|阿塞拜疆文 |Azerbaijani    |az|
+|波斯尼亚文|Bosnian|bs|
+|捷克文|Czech|cs|
+|威尔士文 |Welsh |cy|
+|丹麦文 |Danish|da|
+|爱沙尼亚文 |Estonian |et|
+|爱尔兰文 |Irish |ga|
+|克罗地亚文|Croatian |hr|
+|匈牙利文|Hungarian |hu|
+|印尼文|Indonesian|id|
+|冰岛文 |Icelandic|is|
+|库尔德文 |Kurdish|ku|
+|立陶宛文|Lithuanian |lt|
+|拉脱维亚文 |Latvian |lv|
+|毛利文|Maori|mi|
+|马来文 |Malay|ms|
+|马耳他文 |Maltese |mt|
+|荷兰文 |Dutch |nl|
+|挪威文 |Norwegian |no|
+|波兰文|Polish |pl|
+| 罗马尼亚文|Romanian |ro|
+| 斯洛伐克文|Slovak |sk|
+| 斯洛文尼亚文|Slovenian |sl|
+| 阿尔巴尼亚文|Albanian |sq|
+| 瑞典文|Swedish |sv|
+| 西瓦希里文|Swahili |sw|
+| 塔加洛文|Tagalog |tl|
+| 土耳其文|Turkish |tr|
+| 乌兹别克文|Uzbek |uz|
+| 越南文|Vietnamese |vi|
+| 蒙古文|Mongolian |mn|
+| 阿巴扎文|Abaza |abq|
+| 阿迪赫文|Adyghe |ady|
+| 卡巴丹文|Kabardian |kbd|
+| 阿瓦尔文|Avar |ava|
+| 达尔瓦文|Dargwa |dar|
+| 因古什文|Ingush |inh|
+| 拉克文|Lak |lbe|
+| 莱兹甘文|Lezghian |lez|
+|塔巴萨兰文 |Tabassaran |tab|
+| 比尔哈文|Bihari |bh|
+| 迈蒂利文|Maithili |mai|
+| 昂加文|Angika |ang|
+| 孟加拉文|Bhojpuri |bho|
+| 摩揭陀文 |Magahi |mah|
+| 那格浦尔文|Nagpur |sck|
+| 尼瓦尔文|Newari |new|
+| 保加利亚文 |Goan Konkani|gom|
+| 沙特阿拉伯文|Saudi Arabia|sa|

doc/doc_ch/pgnet.md

+# 端对端OCR算法-PGNet
+- [一、简介](#简介)
+- [二、环境配置](#环境配置)
+- [三、快速使用](#快速使用)
+- [四、模型训练、评估、推理](#模型训练、评估、推理)
+
+<a name="简介"></a>
+## 一、简介
+OCR算法可以分为两阶段算法和端对端的算法。二阶段OCR算法一般分为两个部分，文本检测和文本识别算法，文件检测算法从图像中得到文本行的检测框，然后识别算法去识别文本框中的内容。而端对端OCR算法可以在一个算法中完成文字检测和文字识别，其基本思想是设计一个同时具有检测单元和识别模块的模型，共享其中两者的CNN特征，并联合训练。由于一个算法即可完成文字识别，端对端模型更小，速度更快。
+
+### PGNet算法介绍
+近些年来，端对端OCR算法得到了良好的发展，包括MaskTextSpotter系列、TextSnake、TextDragon、PGNet系列等算法。在这些算法中，PGNet算法具备其他算法不具备的优势，包括：
+- 设计PGNet loss指导训练，不需要字符级别的标注
+- 不需要NMS和ROI相关操作，加速预测
+- 提出预测文本行内的阅读顺序模块；
+- 提出基于图的修正模块（GRM）来进一步提高模型识别性能
+- 精度更高，预测速度更快
+
+PGNet算法细节详见[论文](https://www.aaai.org/AAAI21Papers/AAAI-2885.WangP.pdf) ,算法原理图如下所示：
+![](../pgnet_framework.png)
+输入图像经过特征提取送入四个分支，分别是：文本边缘偏移量预测TBO模块，文本中心线预测TCL模块，文本方向偏移量预测TDO模块，以及文本字符分类图预测TCC模块。
+其中TBO以及TCL的输出经过后处理后可以得到文本的检测结果，TCL、TDO、TCC负责文本识别。
+
+其检测识别效果图如下：
+
+![](../imgs_results/e2e_res_img293_pgnet.png)
+![](../imgs_results/e2e_res_img295_pgnet.png)
+
+### 性能指标
+
+测试集: Total Text
+
+测试环境: NVIDIA Tesla V100-SXM2-16GB
+
+|PGNetA|det_precision|det_recall|det_f_score|e2e_precision|e2e_recall|e2e_f_score|FPS|下载|
+| --- | --- | --- | --- | --- | --- | --- | --- | --- |
+|Paper|85.30|86.80|86.1|-|-|61.7|38.20 (size=640)|-|
+|Ours|87.03|82.48|84.69|61.71|58.43|60.03|48.73 (size=768)|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/en_server_pgnetA.tar)|
+
+*note：PaddleOCR里的PGNet实现针对预测速度做了优化，在精度下降可接受范围内，可以显著提升端对端预测速度*
+
+
+
+<a name="环境配置"></a>
+## 二、环境配置
+请先参考[快速安装](./installation.md)配置PaddleOCR运行环境。
+
+<a name="快速使用"></a>
+## 三、快速使用
+### inference模型下载
+本节以训练好的端到端模型为例，快速使用模型预测，首先下载训练好的端到端inference模型[下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/e2e_server_pgnetA_infer.tar)
+```
+mkdir inference && cd inference
+# 下载英文端到端模型并解压
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/e2e_server_pgnetA_infer.tar && tar xf e2e_server_pgnetA_infer.tar
+```
+* windows 环境下如果没有安装wget,下载模型时可将链接复制到浏览器中下载，并解压放置在相应目录下
+
+解压完毕后应有如下文件结构：
+```
+├── e2e_server_pgnetA_infer
+│   ├── inference.pdiparams
+│   ├── inference.pdiparams.info
+│   └── inference.pdmodel
+```
+### 单张图像或者图像集合预测
+```bash
+# 预测image_dir指定的单张图像
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/img623.jpg" --e2e_model_dir="./inference/e2e_server_pgnetA_infer/" --e2e_pgnet_polygon=True
+
+# 预测image_dir指定的图像集合
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/" --e2e_model_dir="./inference/e2e_server_pgnetA_infer/" --e2e_pgnet_polygon=True
+
+# 如果想使用CPU进行预测，需设置use_gpu参数为False
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/img623.jpg" --e2e_model_dir="./inference/e2e_server_pgnetA_infer/" --e2e_pgnet_polygon=True --use_gpu=False
+```
+### 可视化结果
+可视化文本检测结果默认保存到./inference_results文件夹里面，结果文件的名称前缀为'e2e_res'。结果示例如下：
+![](../imgs_results/e2e_res_img623_pgnet.jpg)
+
+<a name="模型训练、评估、推理"></a>
+## 四、模型训练、评估、推理
+本节以totaltext数据集为例，介绍PaddleOCR中端到端模型的训练、评估与测试。
+
+###  准备数据
+下载解压[totaltext](https://github.com/cs-chan/Total-Text-Dataset/blob/master/Dataset/README.md) 数据集到PaddleOCR/train_data/目录，数据集组织结构：
+```
+/PaddleOCR/train_data/total_text/train/
+  |- rgb/            # total_text数据集的训练数据
+      |- gt_0.png
+      | ...  
+  |- total_text.txt  # total_text数据集的训练标注
+```
+
+total_text.txt标注文件格式如下，文件名和标注信息中间用"\t"分隔：
+```
+" 图像文件名                    json.dumps编码的图像标注信息"
+rgb/gt_0.png    [{"transcription": "EST", "points": [[1004.0,689.0],[1019.0,698.0],[1034.0,708.0],[1049.0,718.0],[1064.0,728.0],[1079.0,738.0],[1095.0,748.0],[1094.0,774.0],[1079.0,765.0],[1065.0,756.0],[1050.0,747.0],[1036.0,738.0],[1021.0,729.0],[1007.0,721.0]]}, {...}]
+```
+json.dumps编码前的图像标注信息是包含多个字典的list，字典中的 `points` 表示文本框的四个点的坐标(x, y)，从左上角的点开始顺时针排列。
+`transcription` 表示当前文本框的文字，**当其内容为“###”时，表示该文本框无效，在训练时会跳过。**
+如果您想在其他数据集上训练，可以按照上述形式构建标注文件。
+
+### 启动训练
+
+PGNet训练分为两个步骤：step1: 在合成数据上训练，得到预训练模型，此时模型精度依然较低；step2: 加载预训练模型，在totaltext数据集上训练；为快速训练，我们直接提供了step1的预训练模型。
+```shell
+cd PaddleOCR/
+下载step1 预训练模型
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/train_step1.tar
+可以得到以下的文件格式
+./pretrain_models/train_step1/
+  └─ best_accuracy.pdopt
+  └─ best_accuracy.states
+  └─ best_accuracy.pdparams
+```
+*如果您安装的是cpu版本，请将配置文件中的 `use_gpu` 字段修改为false*
+
+```shell
+# 单机单卡训练 e2e 模型
+python3 tools/train.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.pretrained_model=./pretrain_models/train_step1/best_accuracy Global.load_static_weights=False
+# 单机多卡训练，通过 --gpus 参数设置使用的GPU ID
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.pretrained_model=./pretrain_models/train_step1/best_accuracy  Global.load_static_weights=False
+```
+
+上述指令中，通过-c 选择训练使用configs/e2e/e2e_r50_vd_pg.yml配置文件。
+有关配置文件的详细解释，请参考[链接](./config.md)。
+
+您也可以通过-o参数在不需要修改yml文件的情况下，改变训练的参数，比如，调整训练的学习率为0.0001
+```shell
+python3 tools/train.py -c configs/e2e/e2e_r50_vd_pg.yml -o Optimizer.base_lr=0.0001
+```
+
+#### 断点训练
+如果训练程序中断，如果希望加载训练中断的模型从而恢复训练，可以通过指定Global.checkpoints指定要加载的模型路径：
+```shell
+python3 tools/train.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.checkpoints=./your/trained/model
+```
+
+**注意**：`Global.checkpoints`的优先级高于`Global.pretrain_weights`的优先级，即同时指定两个参数时，优先加载`Global.checkpoints`指定的模型，如果`Global.checkpoints`指定的模型路径有误，会加载`Global.pretrain_weights`指定的模型。
+
+PaddleOCR计算三个OCR端到端相关的指标，分别是：Precision、Recall、Hmean。
+
+运行如下代码，根据配置文件`e2e_r50_vd_pg.yml`中`save_res_path`指定的测试集检测结果文件，计算评估指标。
+
+评估时设置后处理参数`max_side_len=768`，使用不同数据集、不同模型训练，可调整参数进行优化
+训练中模型参数默认保存在`Global.save_model_dir`目录下。在评估指标时，需要设置`Global.checkpoints`指向保存的参数文件。
+```shell
+python3 tools/eval.py -c configs/e2e/e2e_r50_vd_pg.yml  -o Global.checkpoints="{path/to/weights}/best_accuracy"
+```
+
+### 模型预测
+测试单张图像的端到端识别效果
+```shell
+python3 tools/infer_e2e.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.infer_img="./doc/imgs_en/img_10.jpg" Global.pretrained_model="./output/e2e_pgnet/best_accuracy" Global.load_static_weights=false
+```
+
+测试文件夹下所有图像的端到端识别效果
+```shell
+python3 tools/infer_e2e.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.infer_img="./doc/imgs_en/" Global.pretrained_model="./output/e2e_pgnet/best_accuracy" Global.load_static_weights=false
+```
+
+### 预测推理
+#### (1). 四边形文本检测模型（ICDAR2015）  
+首先将PGNet端到端训练过程中保存的模型，转换成inference model。以基于Resnet50_vd骨干网络，以英文数据集训练的模型为例[模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/en_server_pgnetA.tar) ，可以使用如下命令进行转换：
+```
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/en_server_pgnetA.tar && tar xf en_server_pgnetA.tar
+python3 tools/export_model.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.pretrained_model=./en_server_pgnetA/best_accuracy Global.load_static_weights=False Global.save_inference_dir=./inference/e2e
+```
+**PGNet端到端模型推理，需要设置参数`--e2e_algorithm="PGNet"`**，可以执行如下命令：
+```
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/img_10.jpg" --e2e_model_dir="./inference/e2e/"  --e2e_pgnet_polygon=False
+```
+可视化文本检测结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为'e2e_res'。结果示例如下：
+
+![](../imgs_results/e2e_res_img_10_pgnet.jpg)
+
+#### (2). 弯曲文本检测模型（Total-Text）
+对于弯曲文本样例
+
+**PGNet端到端模型推理，需要设置参数`--e2e_algorithm="PGNet"`，同时，还需要增加参数`--e2e_pgnet_polygon=True`，**可以执行如下命令：
+```
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/img623.jpg" --e2e_model_dir="./inference/e2e/" --e2e_pgnet_polygon=True
+```
+可视化文本端到端结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为'e2e_res'。结果示例如下：
+
+![](../imgs_results/e2e_res_img623_pgnet.jpg)

doc/doc_ch/recognition.md

 ## 文字识别
 
 
-- [一、数据准备](#数据准备)
-    - [数据下载](#数据下载)
-    - [自定义数据集](#自定义数据集)  
-    - [字典](#字典)  
-    - [支持空格](#支持空格)
+- [1 数据准备](#数据准备)
+    - [1.1 自定义数据集](#自定义数据集)
+    - [1.2 数据下载](#数据下载)
+    - [1.3 字典](#字典)  
+    - [1.4 支持空格](#支持空格)
 
-- [二、启动训练](#启动训练)
-    - [1. 数据增强](#数据增强)
-    - [2. 训练](#训练)
-    - [3. 小语种](#小语种)
+- [2 启动训练](#启动训练)
+    - [2.1 数据增强](#数据增强)
+    - [2.2 训练](#训练)
+    - [2.3 小语种](#小语种)
 
-- [三、评估](#评估)
+- [3 评估](#评估)
 
-- [四、预测](#预测)
-    - [1. 训练引擎预测](#训练引擎预测)
+- [4 预测](#预测)
+    - [4.1 训练引擎预测](#训练引擎预测)
 
 
 <a name="数据准备"></a>
-### 数据准备
+### 1. 数据准备
 
 
-PaddleOCR 支持两种数据格式: `lmdb` 用于训练公开数据，调试算法; `通用数据` 训练自己的数据:
-
-请按如下步骤设置数据集：
+PaddleOCR 支持两种数据格式:
+ - `lmdb` 用于训练以lmdb格式存储的数据集;
+ - `通用数据` 用于训练以文本文件存储的数据集:
 
 训练数据的默认存储路径是 `PaddleOCR/train_data`,如果您的磁盘上已有数据集，只需创建软链接至数据集目录：
 
 ```
+# linux and mac os
 ln -sf <path/to/dataset> <path/to/paddle_ocr>/train_data/dataset
+# windows
+mklink /d <path/to/paddle_ocr>/train_data/dataset <path/to/dataset>
 ```
 
-<a name="数据下载"></a>
-* 数据下载
-
-若您本地没有数据集，可以在官网下载 [icdar2015](http://rrc.cvc.uab.es/?ch=4&com=downloads) 数据，用于快速验证。也可以参考[DTRB](https://github.com/clovaai/deep-text-recognition-benchmark#download-lmdb-dataset-for-traininig-and-evaluation-from-here)，下载 benchmark 所需的lmdb格式数据集。
-如果希望复现SRN的论文指标，需要下载离线[增广数据](https://pan.baidu.com/s/1-HSZ-ZVdqBF2HaBZ5pRAKA),提取码: y3ry。增广数据是由MJSynth和SynthText做旋转和扰动得到的。数据下载完成后请解压到 {your_path}/PaddleOCR/train_data/data_lmdb_release/training/ 路径下。
+<a name="准备数据集"></a>
+#### 1.1 自定义数据集
+下面以通用数据集为例， 介绍如何准备数据集：
 
-<a name="自定义数据集"></a>
-* 使用自己数据集
+* 训练集
 
-若您希望使用自己的数据进行训练，请参考下文组织您的数据。
+建议将训练图片放入同一个文件夹，并用一个txt文件（rec_gt_train.txt）记录图片路径和标签，txt文件里的内容如下:
 
-- 训练集
+**注意：** txt文件中默认请将图片路径和图片标签用 \t 分割，如用其他方式分割将造成训练报错。
 
-首先请将训练图片放入同一个文件夹（train_images），并用一个txt文件（rec_gt_train.txt）记录图片路径和标签。
+```
+" 图像文件名                 图像标注信息 "
 
-**注意：** 默认请将图片路径和图片标签用 \t 分割，如用其他方式分割将造成训练报错
+train_data/rec/train/word_001.jpg   简单可依赖
+train_data/rec/train/word_002.jpg   用科技让复杂的世界更简单
+...
+```
 
+最终训练集应有如下文件结构：
+```
+|-train_data
+  |-rec
+    |- rec_gt_train.txt
+    |- train
+        |- word_001.png
+        |- word_002.jpg
+        |- word_003.jpg
+        | ...
 ```
-" 图像文件名                 图像标注信息 "
 
-train_data/train_0001.jpg   简单可依赖
-train_data/train_0002.jpg   用科技让复杂的世界更简单
+- 测试集
+
+同训练集类似，测试集也需要提供一个包含所有图片的文件夹（test）和一个rec_gt_test.txt，测试集的结构如下所示：
+
+```
+|-train_data
+  |-rec
+    |- rec_gt_test.txt
+    |- test
+        |- word_001.jpg
+        |- word_002.jpg
+        |- word_003.jpg
+        | ...
 ```
-PaddleOCR 提供了一份用于训练 icdar2015 数据集的标签文件，通过以下方式下载：
+
+<a name="数据下载"></a>
+
+1.2 数据下载
+
+若您本地没有数据集，可以在官网下载 [icdar2015](http://rrc.cvc.uab.es/?ch=4&com=downloads) 数据，用于快速验证。也可以参考[DTRB](https://github.com/clovaai/deep-text-recognition-benchmark#download-lmdb-dataset-for-traininig-and-evaluation-from-here) ，下载 benchmark 所需的lmdb格式数据集。
+
+如果你使用的是icdar2015的公开数据集，PaddleOCR 提供了一份用于训练 icdar2015 数据集的标签文件，通过以下方式下载：
+
+如果希望复现SRN的论文指标，需要下载离线[增广数据](https://pan.baidu.com/s/1-HSZ-ZVdqBF2HaBZ5pRAKA),提取码: y3ry。增广数据是由MJSynth和SynthText做旋转和扰动得到的。数据下载完成后请解压到 {your_path}/PaddleOCR/train_data/data_lmdb_release/training/ 路径下。
 
 ```
 # 训练集标签
@@ -71,34 +104,8 @@ PaddleOCR 也提供了数据格式转换脚本，可以将官网 label 转换支
 python gen_label.py --mode="rec" --input_path="{path/of/origin/label}" --output_label="rec_gt_label.txt"
 ```
 
-最终训练集应有如下文件结构：
-```
-|-train_data
-    |-ic15_data
-        |- rec_gt_train.txt
-        |- train
-            |- word_001.png
-            |- word_002.jpg
-            |- word_003.jpg
-            | ...
-```
-
-- 测试集
-
-同训练集类似，测试集也需要提供一个包含所有图片的文件夹（test）和一个rec_gt_test.txt，测试集的结构如下所示：
-
-```
-|-train_data
-    |-ic15_data
-        |- rec_gt_test.txt
-        |- test
-            |- word_001.jpg
-            |- word_002.jpg
-            |- word_003.jpg
-            | ...
-```
 <a name="字典"></a>
-- 字典
+1.3 字典
 
 最后需要提供一个字典（{word_dict_name}.txt），使模型在训练时，可以将所有出现的字符映射为字典的索引。
 
@@ -115,6 +122,10 @@ n
 
 word_dict.txt 每行有一个单字，将字符与数字索引映射在一起，“and” 将被映射成 [2 5 1]
 
+* 内置字典
+
+PaddleOCR内置了一部分字典，可以按需使用。
+
 `ppocr/utils/ppocr_keys_v1.txt` 是一个包含6623个字符的中文字典
 
 `ppocr/utils/ic15_dict.txt` 是一个包含36个字符的英文字典
@@ -127,10 +138,10 @@ word_dict.txt 每行有一个单字，将字符与数字索引映射在一起，
 
 `ppocr/utils/dict/german_dict.txt` 是一个包含131个字符的德文字典
 
-`ppocr/utils/dict/en_dict.txt` 是一个包含63个字符的英文字典
+`ppocr/utils/en_dict.txt` 是一个包含96个字符的英文字典
+
 
 
-您可以按需使用。
 
 目前的多语言模型仍处在demo阶段，会持续优化模型并补充语种，**非常欢迎您为我们提供其他语言的字典和字体**，
 如您愿意可将字典文件提交至 [dict](../../ppocr/utils/dict)，我们会在Repo中感谢您。
@@ -141,13 +152,13 @@ word_dict.txt 每行有一个单字，将字符与数字索引映射在一起，
 并将 `character_type` 设置为 `ch`。
 
 <a name="支持空格"></a>
-- 添加空格类别
+1.4 添加空格类别
 
 如果希望支持识别"空格"类别, 请将yml文件中的 `use_space_char` 字段设置为 `True`。
 
 
 <a name="启动训练"></a>
-### 启动训练
+### 2. 启动训练
 
 PaddleOCR提供了训练脚本、评估脚本和预测脚本，本节将以 CRNN 识别模型为例：
 
@@ -172,7 +183,7 @@ tar -xf rec_mv3_none_bilstm_ctc_v2.0_train.tar && rm -rf rec_mv3_none_bilstm_ctc
 python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_icdar15_train.yml
 ```
 <a name="数据增强"></a>
-- 数据增强
+#### 2.1 数据增强
 
 PaddleOCR提供了多种数据增强方式，如果您希望在训练时加入扰动，请在配置文件中设置 `distort: true`。
 
@@ -183,7 +194,7 @@ PaddleOCR提供了多种数据增强方式，如果您希望在训练时加入
 *由于OpenCV的兼容性问题，扰动操作暂时只支持Linux*
 
 <a name="训练"></a>
-- 训练
+#### 2.2 训练
 
 PaddleOCR支持训练和评估交替进行, 可以在 `configs/rec/rec_icdar15_train.yml` 中修改 `eval_batch_step` 设置评估频率，默认每500个iter评估一次。评估过程中默认将最佳acc模型，保存为 `output/rec_CRNN/best_accuracy` 。
 
@@ -272,9 +283,9 @@ Eval:
 **注意，预测/评估时的配置文件请务必与训练一致。**
 
 <a name="小语种"></a>
-- 小语种
+#### 2.3 小语种
 
-PaddleOCR目前已支持26种（除中文外）语种识别，`configs/rec/multi_languages` 路径下提供了一个多语言的配置文件模版: [rec_multi_language_lite_train.yml](../../configs/rec/multi_language/rec_multi_language_lite_train.yml)。
+PaddleOCR目前已支持80种（除中文外）语种识别，`configs/rec/multi_languages` 路径下提供了一个多语言的配置文件模版: [rec_multi_language_lite_train.yml](../../configs/rec/multi_language/rec_multi_language_lite_train.yml)。
 
 您有两种方式创建所需的配置文件：
 
@@ -357,26 +368,12 @@ PaddleOCR目前已支持26种（除中文外）语种识别，`configs/rec/multi
 | rec_ger_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 德语   | german |
 | rec_japan_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 日语  | japan |
 | rec_korean_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 韩语  | korean |
-| rec_it_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 意大利语  | it |
-| rec_xi_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 西班牙语 |  xi |
-| rec_pu_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 葡萄牙语   | pu |
-| rec_ru_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 俄罗斯语  | ru |
-| rec_ar_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 阿拉伯语  | ar |
-| rec_hi_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 印地语 |  hi |
-| rec_ug_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 维吾尔语  | ug |
-| rec_fa_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 波斯语  | fa |
-| rec_ur_ite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 乌尔都语  | ur |
-| rec_rs_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 塞尔维亚(latin)语 | rs |
-| rec_oc_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 欧西坦语   | oc |
-| rec_mr_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 马拉地语  | mr |
-| rec_ne_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 尼泊尔语  | ne |
-| rec_rsc_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 塞尔维亚(cyrillic)语 |  rsc |
-| rec_bg_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 保加利亚语  | bg |
-| rec_uk_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 乌克兰语  | uk |
-| rec_be_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 白俄罗斯语   | be |
-| rec_te_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 泰卢固语  | te |
-| rec_ka_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 卡纳达语  | ka |
-| rec_ta_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 泰米尔语 |  ta |
+| rec_latin_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 拉丁字母  | latin |
+| rec_arabic_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 阿拉伯字母 |  ar |
+| rec_cyrillic_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 斯拉夫字母  | cyrillic |
+| rec_devanagari_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 梵文字母  | devanagari |
+
+更多支持语种请参考: [多语言模型](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.1/doc/doc_ch/multi_languages.md#%E8%AF%AD%E7%A7%8D%E7%BC%A9%E5%86%99)
 
 多语言模型训练方式与中文模型一致，训练数据集均为100w的合成数据，少量的字体可以在 [百度网盘](https://pan.baidu.com/s/1bS_u207Rm7YbY33wOECKDA) 上下载，提取码：frgi。
 
@@ -415,7 +412,7 @@ Eval:
     ...
 ```
 <a name="评估"></a>
-### 评估
+### 3 评估
 
 评估数据集可以通过 `configs/rec/rec_icdar15_train.yml`  修改Eval中的 `label_file_path` 设置。
 
@@ -425,10 +422,10 @@ python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec
 ```
 
 <a name="预测"></a>
-### 预测
+### 4 预测
 
 <a name="训练引擎预测"></a>
-* 训练引擎的预测
+#### 4.1 训练引擎的预测
 
 使用 PaddleOCR 训练好的模型，可以通过以下脚本进行快速预测。