请教一下版面分析+OCR服务部署

[lycfight]

``请提供下述完整信息以便快速定位问题/Please provide the following information to quickly locate the problem

• 系统环境/System Environment：
• 版本号/Version：Paddle： PaddleOCR： 问题相关组件/Related components：
• 运行指令/Command Code：
• 完整报错/Complete Error Message：
  目前尝试了Paddle Hub封装的方法，发现2.0版本后被弃用，只能基于Paddle Serving写服务。不过现有文档看不出如何将一个版面分析+OCR的多阶段pipeline写成服务的教程。样例也只有一个OCR的检测+识别样例：
  
  如果我想要在前面添加版面分析模型，然后再筛选特定块进行OCR的化，如何自定义呢？从给出的样例里是仅需要定义各阶段的OP类吗？需要实现哪些方法呢？从文档和代码里均未发现模型推理的部分，仅有预处理和后处理。

# 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 paddle_serving_server.web_service import WebService, Op

import logging
import numpy as np
import copy
import cv2
import base64
# from paddle_serving_app.reader import OCRReader
from ocr_reader import OCRReader, DetResizeForTest, ArgsParser
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": 0.6,
            "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'))
        self.raw_im = data
        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.ori_h, self.ori_w, _ = im.shape
        det_img = self.det_preprocess(im)
        _, self.new_h, self.new_w = det_img.shape
        return {"x": det_img[np.newaxis, :].copy()}, False, None, ""

    def postprocess(self, input_dicts, fetch_dict, data_id, log_id):
        det_out = list(fetch_dict.values())[0]
        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.raw_im}
        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()
        raw_im = input_dict["image"]
        data = np.frombuffer(raw_im, np.uint8)
        im = cv2.imdecode(data, cv2.IMREAD_COLOR)
        self.dt_list = input_dict["dt_boxes"]
        self.dt_list = self.sorted_boxes(self.dt_list)
        # deepcopy to save origin dt_boxes
        dt_boxes = copy.deepcopy(self.dt_list)
        feed_list = []
        img_list = []
        max_wh_ratio = 320 / 48.
        ## Many mini-batchs, the type of feed_data is list.
        max_batch_size = 6  # len(dt_boxes)

        # If max_batch_size is 0, skipping predict stage
        if max_batch_size == 0:
            return {}, True, None, ""
        boxes_size = len(dt_boxes)
        batch_size = boxes_size // max_batch_size
        rem = boxes_size % max_batch_size
        for bt_idx in range(0, batch_size + 1):
            imgs = None
            boxes_num_in_one_batch = 0
            if bt_idx == batch_size:
                if rem == 0:
                    continue
                else:
                    boxes_num_in_one_batch = rem
            elif bt_idx < batch_size:
                boxes_num_in_one_batch = max_batch_size
            else:
                _LOGGER.error("batch_size error, bt_idx={}, batch_size={}".
                              format(bt_idx, batch_size))
                break

            start = bt_idx * max_batch_size
            end = start + boxes_num_in_one_batch
            img_list = []
            for box_idx in range(start, end):
                boximg = self.get_rotate_crop_image(im, dt_boxes[box_idx])
                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((boxes_num_in_one_batch, 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
            feed = {"x": imgs.copy()}
            feed_list.append(feed)
        return feed_list, False, None, ""

    def postprocess(self, input_dicts, fetch_data, data_id, log_id):
        rec_list = []
        dt_num = len(self.dt_list)
        if isinstance(fetch_data, dict):
            if len(fetch_data) > 0:
                rec_batch_res = self.ocr_reader.postprocess(
                    fetch_data, with_score=True)
                for res in rec_batch_res:
                    rec_list.append(res)
        elif isinstance(fetch_data, list):
            for one_batch in fetch_data:
                one_batch_res = self.ocr_reader.postprocess(
                    one_batch, with_score=True)
                for res in one_batch_res:
                    rec_list.append(res)
        result_list = []
        for i in range(dt_num):
            text = rec_list[i]
            dt_box = self.dt_list[i]
            if text[1] >= 0.5:
                result_list.append([text, dt_box.tolist()])
        res = {"result": str(result_list)}
        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")
FLAGS = ArgsParser().parse_args()
uci_service.prepare_pipeline_config(yml_dict=FLAGS.conf_dict)
uci_service.run_service()

我现有一段推理的代码，先版面分析，然后将OCR的行聚合到文本块中，该如何将其写成服务呢？

import os
import time
import cv2
from paddleocr import PPStructure, PaddleOCR, draw_ocr, draw_structure_result, save_structure_res
from paddleocr.ppstructure.recovery.recovery_to_doc import sorted_layout_boxes

def is_contained(bbox1, bbox2):
    left1, top1, right1, bottom1 = bbox1
    left2, top2, right2, bottom2 = bbox2
    center_x, center_y = (left1 + right1) / 2, (top1 + bottom1) / 2
    if center_x > left2 and center_x < right2 and center_y > top2 and center_y < bottom2:
        return True
    return False

def find_contain_block(ocr_bbox, layout_blocks):
    for block in layout_blocks:
        if is_contained(ocr_bbox['bbox'], block['bbox']):
            return block
    return None
    
def intersection_ratio(bbox1, bbox2):
    left1, top1, right1, bottom1 = bbox1
    left2, top2, right2, bottom2 = bbox2

    left = max(left1, left2)
    top = max(top1, top2)
    right = min(right1, right2)
    bottom = min(bottom1, bottom2)

    if left < right and top < bottom:
        intersection_area = (right - left) * (bottom - top)
        bbox1_area = (right1 - left1) * (bottom1 - top1)
        return intersection_area / bbox1_area
    else:
        return 0.0

def find_max_intersection(ocr_line, layout_blocks):
    max_ratio = 0.0
    max_block = None
    for block in layout_blocks:
        ratio = intersection_ratio(ocr_line['bbox'], block['bbox'])
        if ratio > max_ratio:
            max_ratio = ratio
            max_block = block
    return max_block if max_ratio > 0.9 else None

def aggregate(ocr_results, layout_results):
    layout_blocks = [{'bbox': block['bbox'], 'res': '', 'lines': []} for block in layout_results]
    for line in ocr_results:
        block = find_contain_block(line, layout_blocks)
        if block is not None:
            block['lines'].append(line)
    for block in layout_blocks:
        block['lines'].sort(key=lambda line: line['bbox'][1])
        block['res'] = ''.join([line['res'] for line in block['lines']])
    return layout_blocks
    
start = time.time()

table_engine = PPStructure(show_log=False, ocr=False)
font_path = './fonts/simfang.ttf' # PaddleOCR下提供字体包

img_path = './pdf/JZDB622X-L-05说明书-A1_split/page-22.png'
img = cv2.imread(img_path)
layout_results = table_engine(img)
h, w, _ = img.shape
layout_results = sorted_layout_boxes(layout_results, w)
layout_results_filtered = [block for block in layout_results if block['type'] != 'figure' and block['type'] != 'table']

ocr = PaddleOCR(show_log=False, ocr_order_method='tb-yx')
ocr_results = ocr.ocr(img, cls=True)
ocr_results_convert = [{'bbox': line[0][0] + line[0][2], 'res': line[1][0]} for line in ocr_results[0]]

results = aggregate(ocr_results_convert, layout_results_filtered)

end = time.time()
print('Running time of all pages: %s Seconds'%(end-start))

[SWHL]

可以看一下FastAPI这个工具来部署成服务

[lycfight]

可以看一下FastAPI这个工具来部署成服务

目前还在尝试将PPStructure封装的预处理和后处理拆出来写成op，用Paddle Serving自定义pipeline，如果还是有问题的话，那只能放弃Paddle Serving，教程文档可读性太差了，完全无法实现自己的服务。

FastAPI是能直接复用上面的推理代码，放在服务的特定方法里是吧

[SWHL]

这个可以参考我整理的RapidStructure