paddlex 3.0.0rc1__py3-none-any.whl → 3.0.1__py3-none-any.whl

paddlex/inference/pipelines/layout_parsing/pipeline_v2.py

@@ -15,9 +15,10 @@ from __future__ import annotations
 
 import copy
 import re
-from typing import Any, Dict, Optional, Tuple, Union
+from typing import Any, Dict, List, Optional, Tuple, Union
 
 import numpy as np
+from PIL import Image
 
 from ....utils import logging
 from ....utils.deps import pipeline_requires_extra
@@ -26,18 +27,31 @@ from ...common.reader import ReadImage
 from ...models.object_detection.result import DetResult
 from ...utils.hpi import HPIConfig
 from ...utils.pp_option import PaddlePredictorOption
+from .._parallel import AutoParallelImageSimpleInferencePipeline
 from ..base import BasePipeline
 from ..ocr.result import OCRResult
-from .result_v2 import LayoutParsingResultV2
-from .utils import gather_imgs, get_single_block_parsing_res, get_sub_regions_ocr_res
-
-
-@pipeline_requires_extra("ocr")
-class LayoutParsingPipelineV2(BasePipeline):
+from .result_v2 import LayoutParsingBlock, LayoutParsingRegion, LayoutParsingResultV2
+from .setting import BLOCK_LABEL_MAP, BLOCK_SETTINGS, LINE_SETTINGS, REGION_SETTINGS
+from .utils import (
+    caculate_bbox_area,
+    calculate_minimum_enclosing_bbox,
+    calculate_overlap_ratio,
+    convert_formula_res_to_ocr_format,
+    format_line,
+    gather_imgs,
+    get_bbox_intersection,
+    get_sub_regions_ocr_res,
+    group_boxes_into_lines,
+    remove_overlap_blocks,
+    shrink_supplement_region_bbox,
+    split_boxes_by_projection,
+    update_region_box,
+)
+
+
+class _LayoutParsingPipelineV2(BasePipeline):
     """Layout Parsing Pipeline V2"""
 
-    entities = ["PP-StructureV3"]
-
     def __init__(
         self,
         config: dict,
@@ -53,9 +67,9 @@ class LayoutParsingPipelineV2(BasePipeline):
             device (str, optional): Device to run the predictions on. Defaults to None.
             pp_option (PaddlePredictorOption, optional): PaddlePredictor options. Defaults to None.
             use_hpip (bool, optional): Whether to use the high-performance
-                inference plugin (HPIP). Defaults to False.
+                inference plugin (HPIP) by default. Defaults to False.
             hpi_config (Optional[Union[Dict[str, Any], HPIConfig]], optional):
-                The high-performance inference configuration dictionary.
+                The default high-performance inference configuration dictionary.
                 Defaults to None.
         """
 
@@ -68,8 +82,7 @@ class LayoutParsingPipelineV2(BasePipeline):
 
         self.inintial_predictor(config)
 
-        self.batch_sampler = ImageBatchSampler(batch_size=1)
-
+        self.batch_sampler = ImageBatchSampler(batch_size=config.get("batch_size", 1))
         self.img_reader = ReadImage(format="BGR")
 
     def inintial_predictor(self, config: dict) -> None:
@@ -83,13 +96,20 @@ class LayoutParsingPipelineV2(BasePipeline):
         """
 
         self.use_doc_preprocessor = config.get("use_doc_preprocessor", True)
-        self.use_general_ocr = config.get("use_general_ocr", True)
         self.use_table_recognition = config.get("use_table_recognition", True)
         self.use_seal_recognition = config.get("use_seal_recognition", True)
+        self.use_region_detection = config.get(
+            "use_region_detection",
+            True,
+        )
         self.use_formula_recognition = config.get(
             "use_formula_recognition",
             True,
         )
+        self.use_chart_recognition = config.get(
+            "use_chart_recognition",
+            False,
+        )
 
         if self.use_doc_preprocessor:
             doc_preprocessor_config = config.get("SubPipelines", {}).get(
@@ -101,6 +121,16 @@ class LayoutParsingPipelineV2(BasePipeline):
             self.doc_preprocessor_pipeline = self.create_pipeline(
                 doc_preprocessor_config,
             )
+        if self.use_region_detection:
+            region_detection_config = config.get("SubModules", {}).get(
+                "RegionDetection",
+                {
+                    "model_config_error": "config error for block_region_detection_model!"
+                },
+            )
+            self.region_detection_model = self.create_model(
+                region_detection_config,
+            )
 
         layout_det_config = config.get("SubModules", {}).get(
             "LayoutDetection",
@@ -123,14 +153,13 @@ class LayoutParsingPipelineV2(BasePipeline):
             layout_kwargs["layout_merge_bboxes_mode"] = layout_merge_bboxes_mode
         self.layout_det_model = self.create_model(layout_det_config, **layout_kwargs)
 
-        if self.use_general_ocr or self.use_table_recognition:
-            general_ocr_config = config.get("SubPipelines", {}).get(
-                "GeneralOCR",
-                {"pipeline_config_error": "config error for general_ocr_pipeline!"},
-            )
-            self.general_ocr_pipeline = self.create_pipeline(
-                general_ocr_config,
-            )
+        general_ocr_config = config.get("SubPipelines", {}).get(
+            "GeneralOCR",
+            {"pipeline_config_error": "config error for general_ocr_pipeline!"},
+        )
+        self.general_ocr_pipeline = self.create_pipeline(
+            general_ocr_config,
+        )
 
         if self.use_seal_recognition:
             seal_recognition_config = config.get("SubPipelines", {}).get(
@@ -165,6 +194,17 @@ class LayoutParsingPipelineV2(BasePipeline):
                 formula_recognition_config,
             )
 
+        if self.use_chart_recognition:
+            chart_recognition_config = config.get("SubModules", {}).get(
+                "ChartRecognition",
+                {
+                    "model_config_error": "config error for block_region_detection_model!"
+                },
+            )
+            self.chart_recognition_model = self.create_model(
+                chart_recognition_config,
+            )
+
         return
 
     def get_text_paragraphs_ocr_res(
@@ -209,12 +249,6 @@ class LayoutParsingPipelineV2(BasePipeline):
             )
             return False
 
-        if input_params["use_general_ocr"] and not self.use_general_ocr:
-            logging.error(
-                "Set use_general_ocr, but the models for general OCR are not initialized.",
-            )
-            return False
-
         if input_params["use_seal_recognition"] and not self.use_seal_recognition:
             logging.error(
                 "Set use_seal_recognition, but the models for seal recognition are not initialized.",
@@ -229,159 +263,643 @@ class LayoutParsingPipelineV2(BasePipeline):
 
         return True
 
-    def get_layout_parsing_res(
+    def standardized_data(
         self,
         image: list,
+        region_det_res: DetResult,
         layout_det_res: DetResult,
         overall_ocr_res: OCRResult,
-        table_res_list: list,
-        seal_res_list: list,
         formula_res_list: list,
-        imgs_in_doc: list,
-        text_det_limit_side_len: Optional[int] = None,
-        text_det_limit_type: Optional[str] = None,
-        text_det_thresh: Optional[float] = None,
-        text_det_box_thresh: Optional[float] = None,
-        text_det_unclip_ratio: Optional[float] = None,
-        text_rec_score_thresh: Optional[float] = None,
+        text_rec_model: Any,
+        text_rec_score_thresh: Union[float, None] = None,
     ) -> list:
         """
         Retrieves the layout parsing result based on the layout detection result, OCR result, and other recognition results.
         Args:
             image (list): The input image.
-            layout_det_res (DetResult): The detection result containing the layout information of the document.
-            overall_ocr_res (OCRResult): The overall OCR result containing text information.
-            table_res_list (list): A list of table recognition results.
-            seal_res_list (list): A list of seal recognition results.
+            overall_ocr_res (OCRResult): An object containing the overall OCR results, including detected text boxes and recognized text. The structure is expected to have:
+                - "input_img": The image on which OCR was performed.
+                - "dt_boxes": A list of detected text box coordinates.
+                - "rec_texts": A list of recognized text corresponding to the detected boxes.
+
+            layout_det_res (DetResult): An object containing the layout detection results, including detected layout boxes and their labels. The structure is expected to have:
+                - "boxes": A list of dictionaries with keys "coordinate" for box coordinates and "block_label" for the type of content.
+
+            table_res_list (list): A list of table detection results, where each item is a dictionary containing:
+                - "block_bbox": The bounding box of the table layout.
+                - "pred_html": The predicted HTML representation of the table.
+
             formula_res_list (list): A list of formula recognition results.
-            text_det_limit_side_len (Optional[int], optional): The maximum side length of the text detection region. Defaults to None.
-            text_det_limit_type (Optional[str], optional): The type of limit for the text detection region. Defaults to None.
-            text_det_thresh (Optional[float], optional): The confidence threshold for text detection. Defaults to None.
-            text_det_box_thresh (Optional[float], optional): The confidence threshold for text detection bounding boxes. Defaults to None
-            text_det_unclip_ratio (Optional[float], optional): The unclip ratio for text detection. Defaults to None.
+            text_rec_model (Any): The text recognition model.
             text_rec_score_thresh (Optional[float], optional): The score threshold for text recognition. Defaults to None.
         Returns:
             list: A list of dictionaries representing the layout parsing result.
         """
+
         matched_ocr_dict = {}
-        image = np.array(image)
+        region_to_block_map = {}
+        block_to_ocr_map = {}
         object_boxes = []
         footnote_list = []
-        max_bottom_text_coordinate = 0
+        paragraph_title_list = []
+        bottom_text_y_max = 0
+        max_block_area = 0.0
+        doc_title_num = 0
+
+        base_region_bbox = [65535, 65535, 0, 0]
+        layout_det_res = remove_overlap_blocks(
+            layout_det_res,
+            threshold=0.5,
+            smaller=True,
+        )
 
-        for object_box_idx, box_info in enumerate(layout_det_res["boxes"]):
+        # convert formula_res_list to OCRResult format
+        convert_formula_res_to_ocr_format(formula_res_list, overall_ocr_res)
+
+        # match layout boxes and ocr boxes and get some information for layout_order_config
+        for box_idx, box_info in enumerate(layout_det_res["boxes"]):
             box = box_info["coordinate"]
             label = box_info["label"].lower()
             object_boxes.append(box)
+            _, _, _, y2 = box
+
+            # update the region box and max_block_area according to the layout boxes
+            base_region_bbox = update_region_box(box, base_region_bbox)
+            max_block_area = max(max_block_area, caculate_bbox_area(box))
+
+            # update_layout_order_config_block_index(layout_order_config, label, box_idx)
 
             # set the label of footnote to text, when it is above the text boxes
             if label == "footnote":
-                footnote_list.append(object_box_idx)
-            if label == "text" and box[3] > max_bottom_text_coordinate:
-                max_bottom_text_coordinate = box[3]
+                footnote_list.append(box_idx)
+            elif label == "paragraph_title":
+                paragraph_title_list.append(box_idx)
+            if label == "text":
+                bottom_text_y_max = max(y2, bottom_text_y_max)
+            if label == "doc_title":
+                doc_title_num += 1
 
             if label not in ["formula", "table", "seal"]:
-                _, matched_idxs = get_sub_regions_ocr_res(
+                _, matched_idxes = get_sub_regions_ocr_res(
                     overall_ocr_res, [box], return_match_idx=True
                 )
-                for matched_idx in matched_idxs:
+                block_to_ocr_map[box_idx] = matched_idxes
+                for matched_idx in matched_idxes:
                     if matched_ocr_dict.get(matched_idx, None) is None:
-                        matched_ocr_dict[matched_idx] = [object_box_idx]
+                        matched_ocr_dict[matched_idx] = [box_idx]
                     else:
-                        matched_ocr_dict[matched_idx].append(object_box_idx)
+                        matched_ocr_dict[matched_idx].append(box_idx)
 
+        # fix the footnote label
         for footnote_idx in footnote_list:
             if (
                 layout_det_res["boxes"][footnote_idx]["coordinate"][3]
-                < max_bottom_text_coordinate
+                < bottom_text_y_max
             ):
                 layout_det_res["boxes"][footnote_idx]["label"] = "text"
 
-        already_processed = set()
-        for matched_idx, layout_box_ids in matched_ocr_dict.items():
-            if len(layout_box_ids) <= 1:
-                continue
-
-            # one ocr is matched to multiple layout boxes, split the text into multiple lines
-            for idx in layout_box_ids:
-                if idx in already_processed:
-                    continue
-
-                already_processed.add(idx)
-                wht_im = np.ones(image.shape, dtype=image.dtype) * 255
-                box = object_boxes[idx]
-                x1, y1, x2, y2 = [int(i) for i in box]
-                wht_im[y1:y2, x1:x2, :] = image[y1:y2, x1:x2, :]
-                sub_ocr_res = next(
-                    self.general_ocr_pipeline(
-                        wht_im,
-                        text_det_limit_side_len=text_det_limit_side_len,
-                        text_det_limit_type=text_det_limit_type,
-                        text_det_thresh=text_det_thresh,
-                        text_det_box_thresh=text_det_box_thresh,
-                        text_det_unclip_ratio=text_det_unclip_ratio,
-                        text_rec_score_thresh=text_rec_score_thresh,
+        # check if there is only one paragraph title and without doc_title
+        only_one_paragraph_title = len(paragraph_title_list) == 1 and doc_title_num == 0
+        if only_one_paragraph_title:
+            paragraph_title_block_area = caculate_bbox_area(
+                layout_det_res["boxes"][paragraph_title_list[0]]["coordinate"]
+            )
+            title_area_max_block_threshold = BLOCK_SETTINGS.get(
+                "title_conversion_area_ratio_threshold", 0.3
+            )
+            if (
+                paragraph_title_block_area
+                > max_block_area * title_area_max_block_threshold
+            ):
+                layout_det_res["boxes"][paragraph_title_list[0]]["label"] = "doc_title"
+
+        # Replace the OCR information of the hurdles.
+        for overall_ocr_idx, layout_box_ids in matched_ocr_dict.items():
+            if len(layout_box_ids) > 1:
+                matched_no = 0
+                overall_ocr_box = copy.deepcopy(
+                    overall_ocr_res["rec_boxes"][overall_ocr_idx]
+                )
+                overall_ocr_dt_poly = copy.deepcopy(
+                    overall_ocr_res["dt_polys"][overall_ocr_idx]
+                )
+                for box_idx in layout_box_ids:
+                    layout_box = layout_det_res["boxes"][box_idx]["coordinate"]
+                    crop_box = get_bbox_intersection(overall_ocr_box, layout_box)
+                    for ocr_idx in block_to_ocr_map[box_idx]:
+                        ocr_box = overall_ocr_res["rec_boxes"][ocr_idx]
+                        iou = calculate_overlap_ratio(ocr_box, crop_box, "small")
+                        if iou > 0.8:
+                            overall_ocr_res["rec_texts"][ocr_idx] = ""
+                    x1, y1, x2, y2 = [int(i) for i in crop_box]
+                    crop_img = np.array(image)[y1:y2, x1:x2]
+                    crop_img_rec_res = list(text_rec_model([crop_img]))[0]
+                    crop_img_dt_poly = get_bbox_intersection(
+                        overall_ocr_dt_poly, layout_box, return_format="poly"
+                    )
+                    crop_img_rec_score = crop_img_rec_res["rec_score"]
+                    crop_img_rec_text = crop_img_rec_res["rec_text"]
+                    text_rec_score_thresh = (
+                        text_rec_score_thresh
+                        if text_rec_score_thresh is not None
+                        else (self.general_ocr_pipeline.text_rec_score_thresh)
                     )
+                    if crop_img_rec_score >= text_rec_score_thresh:
+                        matched_no += 1
+                        if matched_no == 1:
+                            # the first matched ocr be replaced by the first matched layout box
+                            overall_ocr_res["dt_polys"][
+                                overall_ocr_idx
+                            ] = crop_img_dt_poly
+                            overall_ocr_res["rec_boxes"][overall_ocr_idx] = crop_box
+                            overall_ocr_res["rec_polys"][
+                                overall_ocr_idx
+                            ] = crop_img_dt_poly
+                            overall_ocr_res["rec_scores"][
+                                overall_ocr_idx
+                            ] = crop_img_rec_score
+                            overall_ocr_res["rec_texts"][
+                                overall_ocr_idx
+                            ] = crop_img_rec_text
+                        else:
+                            # the other matched ocr be appended to the overall ocr result
+                            overall_ocr_res["dt_polys"].append(crop_img_dt_poly)
+                            overall_ocr_res["rec_boxes"] = np.vstack(
+                                (overall_ocr_res["rec_boxes"], crop_box)
+                            )
+                            overall_ocr_res["rec_polys"].append(crop_img_dt_poly)
+                            overall_ocr_res["rec_scores"].append(crop_img_rec_score)
+                            overall_ocr_res["rec_texts"].append(crop_img_rec_text)
+                            overall_ocr_res["rec_labels"].append("text")
+                            block_to_ocr_map[box_idx].remove(overall_ocr_idx)
+                            block_to_ocr_map[box_idx].append(
+                                len(overall_ocr_res["rec_texts"]) - 1
+                            )
+
+        # use layout bbox to do ocr recognition when there is no matched ocr
+        for layout_box_idx, overall_ocr_idxes in block_to_ocr_map.items():
+            has_text = False
+            for idx in overall_ocr_idxes:
+                if overall_ocr_res["rec_texts"][idx] != "":
+                    has_text = True
+                    break
+            if not has_text and layout_det_res["boxes"][layout_box_idx][
+                "label"
+            ] not in BLOCK_LABEL_MAP.get("vision_labels", []):
+                crop_box = layout_det_res["boxes"][layout_box_idx]["coordinate"]
+                x1, y1, x2, y2 = [int(i) for i in crop_box]
+                crop_img = np.array(image)[y1:y2, x1:x2]
+                crop_img_rec_res = next(text_rec_model([crop_img]))
+                crop_img_dt_poly = get_bbox_intersection(
+                    crop_box, crop_box, return_format="poly"
                 )
-                _, matched_idxs = get_sub_regions_ocr_res(
-                    overall_ocr_res, [box], return_match_idx=True
+                crop_img_rec_score = crop_img_rec_res["rec_score"]
+                crop_img_rec_text = crop_img_rec_res["rec_text"]
+                text_rec_score_thresh = (
+                    text_rec_score_thresh
+                    if text_rec_score_thresh is not None
+                    else (self.general_ocr_pipeline.text_rec_score_thresh)
                 )
-                for matched_idx in sorted(matched_idxs, reverse=True):
-                    del overall_ocr_res["dt_polys"][matched_idx]
-                    del overall_ocr_res["rec_texts"][matched_idx]
-                    overall_ocr_res["rec_boxes"] = np.delete(
-                        overall_ocr_res["rec_boxes"], matched_idx, axis=0
+                if crop_img_rec_score >= text_rec_score_thresh:
+                    overall_ocr_res["rec_boxes"] = np.vstack(
+                        (overall_ocr_res["rec_boxes"], crop_box)
+                    )
+                    overall_ocr_res["rec_polys"].append(crop_img_dt_poly)
+                    overall_ocr_res["rec_scores"].append(crop_img_rec_score)
+                    overall_ocr_res["rec_texts"].append(crop_img_rec_text)
+                    overall_ocr_res["rec_labels"].append("text")
+                    block_to_ocr_map[layout_box_idx].append(
+                        len(overall_ocr_res["rec_texts"]) - 1
                     )
-                    del overall_ocr_res["rec_polys"][matched_idx]
-                    del overall_ocr_res["rec_scores"][matched_idx]
 
-                if sub_ocr_res["rec_boxes"].size > 0:
-                    sub_ocr_res["rec_labels"] = ["text"] * len(sub_ocr_res["rec_texts"])
+        # when there is no layout detection result but there is ocr result, convert ocr detection result to layout detection result
+        if len(layout_det_res["boxes"]) == 0 and len(overall_ocr_res["rec_boxes"]) > 0:
+            for idx, ocr_rec_box in enumerate(overall_ocr_res["rec_boxes"]):
+                base_region_bbox = update_region_box(ocr_rec_box, base_region_bbox)
+                layout_det_res["boxes"].append(
+                    {
+                        "label": "text",
+                        "coordinate": ocr_rec_box,
+                        "score": overall_ocr_res["rec_scores"][idx],
+                    }
+                )
+                block_to_ocr_map[idx] = [idx]
 
-                    overall_ocr_res["dt_polys"].extend(sub_ocr_res["dt_polys"])
-                    overall_ocr_res["rec_texts"].extend(sub_ocr_res["rec_texts"])
-                    overall_ocr_res["rec_boxes"] = np.concatenate(
-                        [overall_ocr_res["rec_boxes"], sub_ocr_res["rec_boxes"]], axis=0
+        block_bboxes = [box["coordinate"] for box in layout_det_res["boxes"]]
+        region_det_res["boxes"] = sorted(
+            region_det_res["boxes"],
+            key=lambda item: caculate_bbox_area(item["coordinate"]),
+        )
+        if len(region_det_res["boxes"]) == 0:
+            region_det_res["boxes"] = [
+                {
+                    "coordinate": base_region_bbox,
+                    "label": "SupplementaryRegion",
+                    "score": 1,
+                }
+            ]
+            region_to_block_map[0] = range(len(block_bboxes))
+        else:
+            block_idxes_set = set(range(len(block_bboxes)))
+            # match block to region
+            for region_idx, region_info in enumerate(region_det_res["boxes"]):
+                matched_idxes = []
+                region_to_block_map[region_idx] = []
+                region_bbox = region_info["coordinate"]
+                for block_idx in block_idxes_set:
+                    overlap_ratio = calculate_overlap_ratio(
+                        region_bbox, block_bboxes[block_idx], mode="small"
                     )
-                    overall_ocr_res["rec_polys"].extend(sub_ocr_res["rec_polys"])
-                    overall_ocr_res["rec_scores"].extend(sub_ocr_res["rec_scores"])
-                    overall_ocr_res["rec_labels"].extend(sub_ocr_res["rec_labels"])
-
-        for formula_res in formula_res_list:
-            x_min, y_min, x_max, y_max = list(map(int, formula_res["dt_polys"]))
-            poly_points = [
-                (x_min, y_min),
-                (x_max, y_min),
-                (x_max, y_max),
-                (x_min, y_max),
+                    if overlap_ratio > REGION_SETTINGS.get(
+                        "match_block_overlap_ratio_threshold", 0.8
+                    ):
+                        region_to_block_map[region_idx].append(block_idx)
+                        matched_idxes.append(block_idx)
+                if len(matched_idxes) > 0:
+                    for block_idx in matched_idxes:
+                        block_idxes_set.remove(block_idx)
+                    matched_bboxes = [block_bboxes[idx] for idx in matched_idxes]
+                    new_region_bbox = calculate_minimum_enclosing_bbox(matched_bboxes)
+                    region_det_res["boxes"][region_idx]["coordinate"] = new_region_bbox
+            # Supplement region when there is no matched block
+            if len(block_idxes_set) > 0:
+                while len(block_idxes_set) > 0:
+                    matched_idxes = []
+                    unmatched_bboxes = [block_bboxes[idx] for idx in block_idxes_set]
+                    supplement_region_bbox = calculate_minimum_enclosing_bbox(
+                        unmatched_bboxes
+                    )
+                    # check if the new region bbox is overlapped with other region bbox, if have, then shrink the new region bbox
+                    for region_info in region_det_res["boxes"]:
+                        region_bbox = region_info["coordinate"]
+                        overlap_ratio = calculate_overlap_ratio(
+                            supplement_region_bbox, region_bbox
+                        )
+                        if overlap_ratio > 0:
+                            supplement_region_bbox, matched_idxes = (
+                                shrink_supplement_region_bbox(
+                                    supplement_region_bbox,
+                                    region_bbox,
+                                    image.shape[1],
+                                    image.shape[0],
+                                    block_idxes_set,
+                                    block_bboxes,
+                                )
+                            )
+                    if len(matched_idxes) == 0:
+                        matched_idxes = list(block_idxes_set)
+                    region_idx = len(region_det_res["boxes"])
+                    region_to_block_map[region_idx] = list(matched_idxes)
+                    for block_idx in matched_idxes:
+                        block_idxes_set.remove(block_idx)
+                    region_det_res["boxes"].append(
+                        {
+                            "coordinate": supplement_region_bbox,
+                            "label": "SupplementaryRegion",
+                            "score": 1,
+                        }
+                    )
+
+        region_block_ocr_idx_map = dict(
+            region_to_block_map=region_to_block_map,
+            block_to_ocr_map=block_to_ocr_map,
+        )
+
+        return region_block_ocr_idx_map, region_det_res, layout_det_res
+
+    def sort_line_by_projection(
+        self,
+        line: List[List[Union[List[int], str]]],
+        input_img: np.ndarray,
+        text_rec_model: Any,
+        text_rec_score_thresh: Union[float, None] = None,
+        direction: str = "vertical",
+    ) -> None:
+        """
+        Sort a line of text spans based on their vertical position within the layout bounding box.
+
+        Args:
+            line (list): A list of spans, where each span is a list containing a bounding box and text.
+            input_img (ndarray): The input image used for OCR.
+            general_ocr_pipeline (Any): The general OCR pipeline used for text recognition.
+
+        Returns:
+            list: The sorted line of text spans.
+        """
+        sort_index = 0 if direction == "horizontal" else 1
+        splited_boxes = split_boxes_by_projection(line, direction)
+        splited_lines = []
+        if len(line) != len(splited_boxes):
+            splited_boxes.sort(key=lambda span: span[0][sort_index])
+            for span in splited_boxes:
+                bbox, text, label = span
+                if label == "text":
+                    crop_img = input_img[
+                        int(bbox[1]) : int(bbox[3]),
+                        int(bbox[0]) : int(bbox[2]),
+                    ]
+                    crop_img_rec_res = list(text_rec_model([crop_img]))[0]
+                    crop_img_rec_score = crop_img_rec_res["rec_score"]
+                    crop_img_rec_text = crop_img_rec_res["rec_text"]
+                    text = (
+                        crop_img_rec_text
+                        if crop_img_rec_score >= text_rec_score_thresh
+                        else ""
+                    )
+                    span[1] = text
+
+                splited_lines.append(span)
+        else:
+            splited_lines = line
+
+        return splited_lines
+
+    def get_block_rec_content(
+        self,
+        image: list,
+        ocr_rec_res: dict,
+        block: LayoutParsingBlock,
+        text_rec_model: Any,
+        text_rec_score_thresh: Union[float, None] = None,
+    ) -> str:
+
+        if len(ocr_rec_res["rec_texts"]) == 0:
+            block.content = ""
+            return block
+
+        lines, text_direction, text_line_height = group_boxes_into_lines(
+            ocr_rec_res,
+            LINE_SETTINGS.get("line_height_iou_threshold", 0.8),
+        )
+
+        # format line
+        text_lines = []
+        need_new_line_num = 0
+        # words start coordinate and stop coordinate in the line
+        words_start_index = 0 if text_direction == "horizontal" else 1
+        words_stop_index = words_start_index + 2
+        lines_start_index = 1 if text_direction == "horizontal" else 3
+        line_width_list = []
+
+        if block.label == "reference":
+            rec_boxes = ocr_rec_res["boxes"]
+            block_start_coordinate = min([box[words_start_index] for box in rec_boxes])
+            block_stop_coordinate = max([box[words_stop_index] for box in rec_boxes])
+        else:
+            block_start_coordinate = block.bbox[words_start_index]
+            block_stop_coordinate = block.bbox[words_stop_index]
+
+        for idx, line in enumerate(lines):
+            line.sort(
+                key=lambda span: (
+                    span[0][words_start_index] // 2,
+                    (
+                        span[0][lines_start_index]
+                        if text_direction == "horizontal"
+                        else -span[0][lines_start_index]
+                    ),
+                )
+            )
+
+            line_width = line[-1][0][words_stop_index] - line[0][0][words_start_index]
+            line_width_list.append(line_width)
+            # merge formula and text
+            ocr_labels = [span[2] for span in line]
+            if "formula" in ocr_labels:
+                line = self.sort_line_by_projection(
+                    line, image, text_rec_model, text_rec_score_thresh, text_direction
+                )
+
+            line_text, need_new_line = format_line(
+                line,
+                text_direction,
+                np.max(line_width_list),
+                block_start_coordinate,
+                block_stop_coordinate,
+                line_gap_limit=text_line_height * 1.5,
+                block_label=block.label,
+            )
+            if need_new_line:
+                need_new_line_num += 1
+            if idx == 0:
+                line_start_coordinate = line[0][0][0]
+                block.seg_start_coordinate = line_start_coordinate
+            elif idx == len(lines) - 1:
+                line_end_coordinate = line[-1][0][2]
+                block.seg_end_coordinate = line_end_coordinate
+            text_lines.append(line_text)
+
+        delim = LINE_SETTINGS["delimiter_map"].get(block.label, "")
+        if need_new_line_num > len(text_lines) * 0.5 and delim == "":
+            text_lines = [text.replace("\n", "") for text in text_lines]
+            delim = "\n"
+        content = delim.join(text_lines)
+        block.content = content
+        block.num_of_lines = len(text_lines)
+        block.direction = text_direction
+        block.text_line_height = text_line_height
+        block.text_line_width = np.mean(line_width_list)
+
+        return block
+
+    def get_layout_parsing_blocks(
+        self,
+        image: list,
+        region_block_ocr_idx_map: dict,
+        region_det_res: DetResult,
+        overall_ocr_res: OCRResult,
+        layout_det_res: DetResult,
+        table_res_list: list,
+        seal_res_list: list,
+        chart_res_list: list,
+        text_rec_model: Any,
+        text_rec_score_thresh: Union[float, None] = None,
+    ) -> list:
+        """
+        Extract structured information from OCR and layout detection results.
+
+        Args:
+            image (list): The input image.
+            overall_ocr_res (OCRResult): An object containing the overall OCR results, including detected text boxes and recognized text. The structure is expected to have:
+                - "input_img": The image on which OCR was performed.
+                - "dt_boxes": A list of detected text box coordinates.
+                - "rec_texts": A list of recognized text corresponding to the detected boxes.
+
+            layout_det_res (DetResult): An object containing the layout detection results, including detected layout boxes and their labels. The structure is expected to have:
+                - "boxes": A list of dictionaries with keys "coordinate" for box coordinates and "block_label" for the type of content.
+
+            table_res_list (list): A list of table detection results, where each item is a dictionary containing:
+                - "block_bbox": The bounding box of the table layout.
+                - "pred_html": The predicted HTML representation of the table.
+
+            seal_res_list (List): A list of seal detection results. The details of each item depend on the specific application context.
+            text_rec_model (Any): A model for text recognition.
+            text_rec_score_thresh (Union[float, None]): The minimum score required for a recognized character to be considered valid. If None, use the default value specified during initialization. Default is None.
+
+        Returns:
+            list: A list of structured boxes where each item is a dictionary containing:
+                - "block_label": The label of the content (e.g., 'table', 'chart', 'image').
+                - The label as a key with either table HTML or image data and text.
+                - "block_bbox": The coordinates of the layout box.
+        """
+
+        table_index = 0
+        seal_index = 0
+        chart_index = 0
+        layout_parsing_blocks: List[LayoutParsingBlock] = []
+
+        for box_idx, box_info in enumerate(layout_det_res["boxes"]):
+
+            label = box_info["label"]
+            block_bbox = box_info["coordinate"]
+            rec_res = {"boxes": [], "rec_texts": [], "rec_labels": []}
+
+            block = LayoutParsingBlock(label=label, bbox=block_bbox)
+
+            if label == "table" and len(table_res_list) > 0:
+                block.content = table_res_list[table_index]["pred_html"]
+                table_index += 1
+            elif label == "seal" and len(seal_res_list) > 0:
+                block.content = "\n".join(seal_res_list[seal_index]["rec_texts"])
+                seal_index += 1
+            elif label == "chart" and len(chart_res_list) > 0:
+                block.content = chart_res_list[chart_index]
+                chart_index += 1
+            else:
+                if label == "formula":
+                    _, ocr_idx_list = get_sub_regions_ocr_res(
+                        overall_ocr_res, [block_bbox], return_match_idx=True
+                    )
+                    region_block_ocr_idx_map["block_to_ocr_map"][box_idx] = ocr_idx_list
+                else:
+                    ocr_idx_list = region_block_ocr_idx_map["block_to_ocr_map"].get(
+                        box_idx, []
+                    )
+                for box_no in ocr_idx_list:
+                    rec_res["boxes"].append(overall_ocr_res["rec_boxes"][box_no])
+                    rec_res["rec_texts"].append(
+                        overall_ocr_res["rec_texts"][box_no],
+                    )
+                    rec_res["rec_labels"].append(
+                        overall_ocr_res["rec_labels"][box_no],
+                    )
+                block = self.get_block_rec_content(
+                    image=image,
+                    block=block,
+                    ocr_rec_res=rec_res,
+                    text_rec_model=text_rec_model,
+                    text_rec_score_thresh=text_rec_score_thresh,
+                )
+
+            if (
+                label
+                in ["seal", "table", "formula", "chart"]
+                + BLOCK_LABEL_MAP["image_labels"]
+            ):
+                x_min, y_min, x_max, y_max = list(map(int, block_bbox))
+                img_path = (
+                    f"imgs/img_in_{block.label}_box_{x_min}_{y_min}_{x_max}_{y_max}.jpg"
+                )
+                img = Image.fromarray(image[y_min:y_max, x_min:x_max, ::-1])
+                block.image = {"path": img_path, "img": img}
+
+            layout_parsing_blocks.append(block)
+
+        region_list: List[LayoutParsingRegion] = []
+        for region_idx, region_info in enumerate(region_det_res["boxes"]):
+            region_bbox = region_info["coordinate"]
+            region_blocks = [
+                layout_parsing_blocks[idx]
+                for idx in region_block_ocr_idx_map["region_to_block_map"][region_idx]
             ]
-            overall_ocr_res["dt_polys"].append(poly_points)
-            overall_ocr_res["rec_texts"].append(f"${formula_res['rec_formula']}$")
-            overall_ocr_res["rec_boxes"] = np.vstack(
-                (overall_ocr_res["rec_boxes"], [formula_res["dt_polys"]])
+            region = LayoutParsingRegion(
+                bbox=region_bbox,
+                blocks=region_blocks,
+                image_shape=image.shape[:2],
             )
-            overall_ocr_res["rec_labels"].append("formula")
-            overall_ocr_res["rec_polys"].append(poly_points)
-            overall_ocr_res["rec_scores"].append(1)
+            region_list.append(region)
+
+        region_list = sorted(
+            region_list,
+            key=lambda r: (r.weighted_distance),
+        )
+
+        return region_list
 
-        parsing_res_list = get_single_block_parsing_res(
-            self.general_ocr_pipeline,
+    def get_layout_parsing_res(
+        self,
+        image: list,
+        region_det_res: DetResult,
+        layout_det_res: DetResult,
+        overall_ocr_res: OCRResult,
+        table_res_list: list,
+        seal_res_list: list,
+        chart_res_list: list,
+        formula_res_list: list,
+        text_rec_score_thresh: Union[float, None] = None,
+    ) -> list:
+        """
+        Retrieves the layout parsing result based on the layout detection result, OCR result, and other recognition results.
+        Args:
+            image (list): The input image.
+            layout_det_res (DetResult): The detection result containing the layout information of the document.
+            overall_ocr_res (OCRResult): The overall OCR result containing text information.
+            table_res_list (list): A list of table recognition results.
+            seal_res_list (list): A list of seal recognition results.
+            formula_res_list (list): A list of formula recognition results.
+            text_rec_score_thresh (Optional[float], optional): The score threshold for text recognition. Defaults to None.
+        Returns:
+            list: A list of dictionaries representing the layout parsing result.
+        """
+
+        # Standardize data
+        region_block_ocr_idx_map, region_det_res, layout_det_res = (
+            self.standardized_data(
+                image=image,
+                region_det_res=region_det_res,
+                layout_det_res=layout_det_res,
+                overall_ocr_res=overall_ocr_res,
+                formula_res_list=formula_res_list,
+                text_rec_model=self.general_ocr_pipeline.text_rec_model,
+                text_rec_score_thresh=text_rec_score_thresh,
+            )
+        )
+
+        # Format layout parsing block
+        region_list = self.get_layout_parsing_blocks(
+            image=image,
+            region_block_ocr_idx_map=region_block_ocr_idx_map,
+            region_det_res=region_det_res,
             overall_ocr_res=overall_ocr_res,
             layout_det_res=layout_det_res,
             table_res_list=table_res_list,
             seal_res_list=seal_res_list,
+            chart_res_list=chart_res_list,
+            text_rec_model=self.general_ocr_pipeline.text_rec_model,
+            text_rec_score_thresh=self.general_ocr_pipeline.text_rec_score_thresh,
         )
 
+        parsing_res_list = []
+        for region in region_list:
+            parsing_res_list.extend(region.sort())
+
+        index = 1
+        for block in parsing_res_list:
+            if block.label in BLOCK_LABEL_MAP["visualize_index_labels"]:
+                block.order_index = index
+                index += 1
+
         return parsing_res_list
 
     def get_model_settings(
         self,
         use_doc_orientation_classify: Union[bool, None],
         use_doc_unwarping: Union[bool, None],
-        use_general_ocr: Union[bool, None],
         use_seal_recognition: Union[bool, None],
         use_table_recognition: Union[bool, None],
         use_formula_recognition: Union[bool, None],
+        use_chart_recognition: Union[bool, None],
+        use_region_detection: Union[bool, None],
     ) -> dict:
         """
         Get the model settings based on the provided parameters or default values.
@@ -389,7 +907,6 @@ class LayoutParsingPipelineV2(BasePipeline):
         Args:
             use_doc_orientation_classify (Union[bool, None]): Enables document orientation classification if True. Defaults to system setting if None.
             use_doc_unwarping (Union[bool, None]): Enables document unwarping if True. Defaults to system setting if None.
-            use_general_ocr (Union[bool, None]): Enables general OCR if True. Defaults to system setting if None.
             use_seal_recognition (Union[bool, None]): Enables seal recognition if True. Defaults to system setting if None.
             use_table_recognition (Union[bool, None]): Enables table recognition if True. Defaults to system setting if None.
             use_formula_recognition (Union[bool, None]): Enables formula recognition if True. Defaults to system setting if None.
@@ -406,9 +923,6 @@ class LayoutParsingPipelineV2(BasePipeline):
             else:
                 use_doc_preprocessor = False
 
-        if use_general_ocr is None:
-            use_general_ocr = self.use_general_ocr
-
         if use_seal_recognition is None:
             use_seal_recognition = self.use_seal_recognition
 
@@ -418,24 +932,32 @@ class LayoutParsingPipelineV2(BasePipeline):
         if use_formula_recognition is None:
             use_formula_recognition = self.use_formula_recognition
 
+        if use_region_detection is None:
+            use_region_detection = self.use_region_detection
+
+        if use_chart_recognition is None:
+            use_chart_recognition = self.use_chart_recognition
+
         return dict(
             use_doc_preprocessor=use_doc_preprocessor,
-            use_general_ocr=use_general_ocr,
             use_seal_recognition=use_seal_recognition,
             use_table_recognition=use_table_recognition,
             use_formula_recognition=use_formula_recognition,
+            use_chart_recognition=use_chart_recognition,
+            use_region_detection=use_region_detection,
         )
 
     def predict(
         self,
         input: Union[str, list[str], np.ndarray, list[np.ndarray]],
-        use_doc_orientation_classify: Union[bool, None] = None,
-        use_doc_unwarping: Union[bool, None] = None,
+        use_doc_orientation_classify: Union[bool, None] = False,
+        use_doc_unwarping: Union[bool, None] = False,
         use_textline_orientation: Optional[bool] = None,
-        use_general_ocr: Union[bool, None] = None,
         use_seal_recognition: Union[bool, None] = None,
         use_table_recognition: Union[bool, None] = None,
         use_formula_recognition: Union[bool, None] = None,
+        use_chart_recognition: Union[bool, None] = False,
+        use_region_detection: Union[bool, None] = None,
         layout_threshold: Optional[Union[float, dict]] = None,
         layout_nms: Optional[bool] = None,
         layout_unclip_ratio: Optional[Union[float, Tuple[float, float], dict]] = None,
@@ -452,7 +974,10 @@ class LayoutParsingPipelineV2(BasePipeline):
         seal_det_box_thresh: Union[float, None] = None,
         seal_det_unclip_ratio: Union[float, None] = None,
         seal_rec_score_thresh: Union[float, None] = None,
-        use_table_cells_ocr_results: bool = False,
+        use_wired_table_cells_trans_to_html: bool = False,
+        use_wireless_table_cells_trans_to_html: bool = False,
+        use_table_orientation_classify: bool = True,
+        use_ocr_results_with_table_cells: bool = True,
         use_e2e_wired_table_rec_model: bool = False,
         use_e2e_wireless_table_rec_model: bool = True,
         **kwargs,
@@ -464,10 +989,10 @@ class LayoutParsingPipelineV2(BasePipeline):
             use_doc_orientation_classify (Optional[bool]): Whether to use document orientation classification.
             use_doc_unwarping (Optional[bool]): Whether to use document unwarping.
             use_textline_orientation (Optional[bool]): Whether to use textline orientation prediction.
-            use_general_ocr (Optional[bool]): Whether to use general OCR.
             use_seal_recognition (Optional[bool]): Whether to use seal recognition.
             use_table_recognition (Optional[bool]): Whether to use table recognition.
             use_formula_recognition (Optional[bool]): Whether to use formula recognition.
+            use_region_detection (Optional[bool]): Whether to use region detection.
             layout_threshold (Optional[float]): The threshold value to filter out low-confidence predictions. Default is None.
             layout_nms (bool, optional): Whether to use layout-aware NMS. Defaults to False.
             layout_unclip_ratio (Optional[Union[float, Tuple[float, float]]], optional): The ratio of unclipping the bounding box.
@@ -488,7 +1013,10 @@ class LayoutParsingPipelineV2(BasePipeline):
             seal_det_box_thresh (Optional[float]): Threshold for seal detection boxes.
             seal_det_unclip_ratio (Optional[float]): Ratio for unclipping seal detection boxes.
             seal_rec_score_thresh (Optional[float]): Score threshold for seal recognition.
-            use_table_cells_ocr_results (bool): whether to use OCR results with cells.
+            use_wired_table_cells_trans_to_html (bool): Whether to use wired table cells trans to HTML.
+            use_wireless_table_cells_trans_to_html (bool): Whether to use wireless table cells trans to HTML.
+            use_table_orientation_classify (bool): Whether to use table orientation classification.
+            use_ocr_results_with_table_cells (bool): Whether to use OCR results processed by table cells.
             use_e2e_wired_table_rec_model (bool): Whether to use end-to-end wired table recognition model.
             use_e2e_wireless_table_rec_model (bool): Whether to use end-to-end wireless table recognition model.
             **kwargs (Any): Additional settings to extend functionality.
@@ -500,150 +1028,204 @@ class LayoutParsingPipelineV2(BasePipeline):
         model_settings = self.get_model_settings(
             use_doc_orientation_classify,
             use_doc_unwarping,
-            use_general_ocr,
             use_seal_recognition,
             use_table_recognition,
             use_formula_recognition,
+            use_chart_recognition,
+            use_region_detection,
         )
 
         if not self.check_model_settings_valid(model_settings):
             yield {"error": "the input params for model settings are invalid!"}
 
         for batch_data in self.batch_sampler(input):
-            image_array = self.img_reader(batch_data.instances)[0]
+            image_arrays = self.img_reader(batch_data.instances)
 
             if model_settings["use_doc_preprocessor"]:
-                doc_preprocessor_res = next(
+                doc_preprocessor_results = list(
                     self.doc_preprocessor_pipeline(
-                        image_array,
+                        image_arrays,
                         use_doc_orientation_classify=use_doc_orientation_classify,
                         use_doc_unwarping=use_doc_unwarping,
-                    ),
+                    )
                 )
             else:
-                doc_preprocessor_res = {"output_img": image_array}
+                doc_preprocessor_results = [{"output_img": arr} for arr in image_arrays]
 
-            doc_preprocessor_image = doc_preprocessor_res["output_img"]
+            doc_preprocessor_images = [
+                item["output_img"] for item in doc_preprocessor_results
+            ]
 
-            layout_det_res = next(
+            layout_det_results = list(
                 self.layout_det_model(
-                    doc_preprocessor_image,
+                    doc_preprocessor_images,
                     threshold=layout_threshold,
                     layout_nms=layout_nms,
                     layout_unclip_ratio=layout_unclip_ratio,
                     layout_merge_bboxes_mode=layout_merge_bboxes_mode,
                 )
             )
-            imgs_in_doc = gather_imgs(doc_preprocessor_image, layout_det_res["boxes"])
+            imgs_in_doc = [
+                gather_imgs(img, res["boxes"])
+                for img, res in zip(doc_preprocessor_images, layout_det_results)
+            ]
+
+            if model_settings["use_region_detection"]:
+                region_det_results = list(
+                    self.region_detection_model(
+                        doc_preprocessor_images,
+                        layout_nms=True,
+                        layout_merge_bboxes_mode="small",
+                    ),
+                )
+            else:
+                region_det_results = [{"boxes": []} for _ in doc_preprocessor_images]
 
             if model_settings["use_formula_recognition"]:
-                formula_res_all = next(
+                formula_res_all = list(
                     self.formula_recognition_pipeline(
-                        doc_preprocessor_image,
+                        doc_preprocessor_images,
                         use_layout_detection=False,
                         use_doc_orientation_classify=False,
                         use_doc_unwarping=False,
-                        layout_det_res=layout_det_res,
+                        layout_det_res=layout_det_results,
                     ),
                 )
-                formula_res_list = formula_res_all["formula_res_list"]
+                formula_res_lists = [
+                    item["formula_res_list"] for item in formula_res_all
+                ]
             else:
-                formula_res_list = []
-
-            for formula_res in formula_res_list:
-                x_min, y_min, x_max, y_max = list(map(int, formula_res["dt_polys"]))
-                doc_preprocessor_image[y_min:y_max, x_min:x_max, :] = 255.0
+                formula_res_lists = [[] for _ in doc_preprocessor_images]
 
-            if (
-                model_settings["use_general_ocr"]
-                or model_settings["use_table_recognition"]
+            for doc_preprocessor_image, formula_res_list in zip(
+                doc_preprocessor_images, formula_res_lists
             ):
-                overall_ocr_res = next(
-                    self.general_ocr_pipeline(
-                        doc_preprocessor_image,
-                        use_textline_orientation=use_textline_orientation,
-                        text_det_limit_side_len=text_det_limit_side_len,
-                        text_det_limit_type=text_det_limit_type,
-                        text_det_thresh=text_det_thresh,
-                        text_det_box_thresh=text_det_box_thresh,
-                        text_det_unclip_ratio=text_det_unclip_ratio,
-                        text_rec_score_thresh=text_rec_score_thresh,
-                    ),
-                )
-            else:
-                overall_ocr_res = {}
+                for formula_res in formula_res_list:
+                    x_min, y_min, x_max, y_max = list(map(int, formula_res["dt_polys"]))
+                    doc_preprocessor_image[y_min:y_max, x_min:x_max, :] = 255.0
+
+            overall_ocr_results = list(
+                self.general_ocr_pipeline(
+                    doc_preprocessor_images,
+                    use_textline_orientation=use_textline_orientation,
+                    text_det_limit_side_len=text_det_limit_side_len,
+                    text_det_limit_type=text_det_limit_type,
+                    text_det_thresh=text_det_thresh,
+                    text_det_box_thresh=text_det_box_thresh,
+                    text_det_unclip_ratio=text_det_unclip_ratio,
+                    text_rec_score_thresh=text_rec_score_thresh,
+                ),
+            )
 
-            overall_ocr_res["rec_labels"] = ["text"] * len(overall_ocr_res["rec_texts"])
+            for overall_ocr_res in overall_ocr_results:
+                overall_ocr_res["rec_labels"] = ["text"] * len(
+                    overall_ocr_res["rec_texts"]
+                )
 
             if model_settings["use_table_recognition"]:
-                table_contents = copy.deepcopy(overall_ocr_res)
-                for formula_res in formula_res_list:
-                    x_min, y_min, x_max, y_max = list(map(int, formula_res["dt_polys"]))
-                    poly_points = [
-                        (x_min, y_min),
-                        (x_max, y_min),
-                        (x_max, y_max),
-                        (x_min, y_max),
-                    ]
-                    table_contents["dt_polys"].append(poly_points)
-                    table_contents["rec_texts"].append(
-                        f"${formula_res['rec_formula']}$"
-                    )
-                    table_contents["rec_boxes"] = np.vstack(
-                        (table_contents["rec_boxes"], [formula_res["dt_polys"]])
+                table_res_lists = []
+                for (
+                    layout_det_res,
+                    doc_preprocessor_image,
+                    overall_ocr_res,
+                    formula_res_list,
+                    imgs_in_doc_for_img,
+                ) in zip(
+                    layout_det_results,
+                    doc_preprocessor_images,
+                    overall_ocr_results,
+                    formula_res_lists,
+                    imgs_in_doc,
+                ):
+                    table_contents_for_img = copy.deepcopy(overall_ocr_res)
+                    for formula_res in formula_res_list:
+                        x_min, y_min, x_max, y_max = list(
+                            map(int, formula_res["dt_polys"])
+                        )
+                        poly_points = [
+                            (x_min, y_min),
+                            (x_max, y_min),
+                            (x_max, y_max),
+                            (x_min, y_max),
+                        ]
+                        table_contents_for_img["dt_polys"].append(poly_points)
+                        rec_formula = formula_res["rec_formula"]
+                        if not rec_formula.startswith("$") or not rec_formula.endswith(
+                            "$"
+                        ):
+                            rec_formula = f"${rec_formula}$"
+                        table_contents_for_img["rec_texts"].append(f"{rec_formula}")
+                        if table_contents_for_img["rec_boxes"].size == 0:
+                            table_contents_for_img["rec_boxes"] = np.array(
+                                [formula_res["dt_polys"]]
+                            )
+                        else:
+                            table_contents_for_img["rec_boxes"] = np.vstack(
+                                (
+                                    table_contents_for_img["rec_boxes"],
+                                    [formula_res["dt_polys"]],
+                                )
+                            )
+                        table_contents_for_img["rec_polys"].append(poly_points)
+                        table_contents_for_img["rec_scores"].append(1)
+
+                    for img in imgs_in_doc_for_img:
+                        img_path = img["path"]
+                        x_min, y_min, x_max, y_max = img["coordinate"]
+                        poly_points = [
+                            (x_min, y_min),
+                            (x_max, y_min),
+                            (x_max, y_max),
+                            (x_min, y_max),
+                        ]
+                        table_contents_for_img["dt_polys"].append(poly_points)
+                        table_contents_for_img["rec_texts"].append(
+                            f'<div style="text-align: center;"><img src="{img_path}" alt="Image" /></div>'
+                        )
+                        if table_contents_for_img["rec_boxes"].size == 0:
+                            table_contents_for_img["rec_boxes"] = np.array(
+                                [img["coordinate"]]
+                            )
+                        else:
+                            table_contents_for_img["rec_boxes"] = np.vstack(
+                                (table_contents_for_img["rec_boxes"], img["coordinate"])
+                            )
+                        table_contents_for_img["rec_polys"].append(poly_points)
+                        table_contents_for_img["rec_scores"].append(img["score"])
+
+                    table_res_all = list(
+                        self.table_recognition_pipeline(
+                            doc_preprocessor_image,
+                            use_doc_orientation_classify=False,
+                            use_doc_unwarping=False,
+                            use_layout_detection=False,
+                            use_ocr_model=False,
+                            overall_ocr_res=table_contents_for_img,
+                            layout_det_res=layout_det_res,
+                            cell_sort_by_y_projection=True,
+                            use_wired_table_cells_trans_to_html=use_wired_table_cells_trans_to_html,
+                            use_wireless_table_cells_trans_to_html=use_wireless_table_cells_trans_to_html,
+                            use_table_orientation_classify=use_table_orientation_classify,
+                            use_ocr_results_with_table_cells=use_ocr_results_with_table_cells,
+                            use_e2e_wired_table_rec_model=use_e2e_wired_table_rec_model,
+                            use_e2e_wireless_table_rec_model=use_e2e_wireless_table_rec_model,
+                        ),
                     )
-                    table_contents["rec_polys"].append(poly_points)
-                    table_contents["rec_scores"].append(1)
-
-                for img in imgs_in_doc:
-                    img_path = img["path"]
-                    x_min, y_min, x_max, y_max = img["coordinate"]
-                    poly_points = [
-                        (x_min, y_min),
-                        (x_max, y_min),
-                        (x_max, y_max),
-                        (x_min, y_max),
+                    single_table_res_lists = [
+                        item["table_res_list"] for item in table_res_all
                     ]
-                    table_contents["dt_polys"].append(poly_points)
-                    table_contents["rec_texts"].append(
-                        f'<div style="text-align: center;"><img src="{img_path}" alt="Image" /></div>'
-                    )
-                    if table_contents["rec_boxes"].size == 0:
-                        table_contents["rec_boxes"] = np.array([img["coordinate"]])
-                    else:
-                        table_contents["rec_boxes"] = np.vstack(
-                            (table_contents["rec_boxes"], img["coordinate"])
-                        )
-                    table_contents["rec_polys"].append(poly_points)
-                    table_contents["rec_scores"].append(img["score"])
-
-                table_res_all = next(
-                    self.table_recognition_pipeline(
-                        doc_preprocessor_image,
-                        use_doc_orientation_classify=False,
-                        use_doc_unwarping=False,
-                        use_layout_detection=False,
-                        use_ocr_model=False,
-                        overall_ocr_res=table_contents,
-                        layout_det_res=layout_det_res,
-                        cell_sort_by_y_projection=True,
-                        use_table_cells_ocr_results=use_table_cells_ocr_results,
-                        use_e2e_wired_table_rec_model=use_e2e_wired_table_rec_model,
-                        use_e2e_wireless_table_rec_model=use_e2e_wireless_table_rec_model,
-                    ),
-                )
-                table_res_list = table_res_all["table_res_list"]
+                    table_res_lists.extend(single_table_res_lists)
             else:
-                table_res_list = []
+                table_res_lists = [[] for _ in doc_preprocessor_images]
 
             if model_settings["use_seal_recognition"]:
-                seal_res_all = next(
+                seal_res_all = list(
                     self.seal_recognition_pipeline(
-                        doc_preprocessor_image,
+                        doc_preprocessor_images,
                         use_doc_orientation_classify=False,
                         use_doc_unwarping=False,
                         use_layout_detection=False,
-                        layout_det_res=layout_det_res,
+                        layout_det_res=layout_det_results,
                         seal_det_limit_side_len=seal_det_limit_side_len,
                         seal_det_limit_type=seal_det_limit_type,
                         seal_det_thresh=seal_det_thresh,
@@ -652,46 +1234,85 @@ class LayoutParsingPipelineV2(BasePipeline):
                         seal_rec_score_thresh=seal_rec_score_thresh,
                     ),
                 )
-                seal_res_list = seal_res_all["seal_res_list"]
+                seal_res_lists = [item["seal_res_list"] for item in seal_res_all]
             else:
-                seal_res_list = []
+                seal_res_lists = [[] for _ in doc_preprocessor_images]
 
-            parsing_res_list = self.get_layout_parsing_res(
+            for (
+                input_path,
+                page_index,
                 doc_preprocessor_image,
-                layout_det_res=layout_det_res,
-                overall_ocr_res=overall_ocr_res,
-                table_res_list=table_res_list,
-                seal_res_list=seal_res_list,
-                formula_res_list=formula_res_list,
-                imgs_in_doc=imgs_in_doc,
-                text_det_limit_side_len=text_det_limit_side_len,
-                text_det_limit_type=text_det_limit_type,
-                text_det_thresh=text_det_thresh,
-                text_det_box_thresh=text_det_box_thresh,
-                text_det_unclip_ratio=text_det_unclip_ratio,
-                text_rec_score_thresh=text_rec_score_thresh,
-            )
+                doc_preprocessor_res,
+                layout_det_res,
+                region_det_res,
+                overall_ocr_res,
+                table_res_list,
+                seal_res_list,
+                formula_res_list,
+                imgs_in_doc_for_img,
+            ) in zip(
+                batch_data.input_paths,
+                batch_data.page_indexes,
+                doc_preprocessor_images,
+                doc_preprocessor_results,
+                layout_det_results,
+                region_det_results,
+                overall_ocr_results,
+                table_res_lists,
+                seal_res_lists,
+                formula_res_lists,
+                imgs_in_doc,
+            ):
+                chart_res_list = []
+                if model_settings["use_chart_recognition"]:
+                    chart_imgs_list = []
+                    for bbox in layout_det_res["boxes"]:
+                        if bbox["label"] == "chart":
+                            x_min, y_min, x_max, y_max = bbox["coordinate"]
+                            chart_img = doc_preprocessor_image[
+                                int(y_min) : int(y_max), int(x_min) : int(x_max), :
+                            ]
+                            chart_imgs_list.append({"image": chart_img})
+
+                    for chart_res_batch in self.chart_recognition_model(
+                        input=chart_imgs_list
+                    ):
+                        chart_res_list.append(chart_res_batch["result"])
+
+                parsing_res_list = self.get_layout_parsing_res(
+                    doc_preprocessor_image,
+                    region_det_res=region_det_res,
+                    layout_det_res=layout_det_res,
+                    overall_ocr_res=overall_ocr_res,
+                    table_res_list=table_res_list,
+                    seal_res_list=seal_res_list,
+                    chart_res_list=chart_res_list,
+                    formula_res_list=formula_res_list,
+                    text_rec_score_thresh=text_rec_score_thresh,
+                )
 
-            for formula_res in formula_res_list:
-                x_min, y_min, x_max, y_max = list(map(int, formula_res["dt_polys"]))
-                doc_preprocessor_image[y_min:y_max, x_min:x_max, :] = formula_res[
-                    "input_img"
-                ]
+                for formula_res in formula_res_list:
+                    x_min, y_min, x_max, y_max = list(map(int, formula_res["dt_polys"]))
+                    doc_preprocessor_image[y_min:y_max, x_min:x_max, :] = formula_res[
+                        "input_img"
+                    ]
 
-            single_img_res = {
-                "input_path": batch_data.input_paths[0],
-                "page_index": batch_data.page_indexes[0],
-                "doc_preprocessor_res": doc_preprocessor_res,
-                "layout_det_res": layout_det_res,
-                "overall_ocr_res": overall_ocr_res,
-                "table_res_list": table_res_list,
-                "seal_res_list": seal_res_list,
-                "formula_res_list": formula_res_list,
-                "parsing_res_list": parsing_res_list,
-                "imgs_in_doc": imgs_in_doc,
-                "model_settings": model_settings,
-            }
-            yield LayoutParsingResultV2(single_img_res)
+                single_img_res = {
+                    "input_path": input_path,
+                    "page_index": page_index,
+                    "doc_preprocessor_res": doc_preprocessor_res,
+                    "layout_det_res": layout_det_res,
+                    "region_det_res": region_det_res,
+                    "overall_ocr_res": overall_ocr_res,
+                    "table_res_list": table_res_list,
+                    "seal_res_list": seal_res_list,
+                    "chart_res_list": chart_res_list,
+                    "formula_res_list": formula_res_list,
+                    "parsing_res_list": parsing_res_list,
+                    "imgs_in_doc": imgs_in_doc_for_img,
+                    "model_settings": model_settings,
+                }
+                yield LayoutParsingResultV2(single_img_res)
 
     def concatenate_markdown_pages(self, markdown_list: list) -> tuple:
         """
@@ -747,3 +1368,15 @@ class LayoutParsingPipelineV2(BasePipeline):
             )
 
         return markdown_texts
+
+
+@pipeline_requires_extra("ocr")
+class LayoutParsingPipelineV2(AutoParallelImageSimpleInferencePipeline):
+    entities = ["PP-StructureV3"]
+
+    @property
+    def _pipeline_cls(self):
+        return _LayoutParsingPipelineV2
+
+    def _get_batch_size(self, config):
+        return config.get("batch_size", 1)