Merge pull request #4965 from opendatalab/dev

3.1.13

Author: myhloli

demo/demo.py

@@ -50,13 +50,15 @@ def build_form_data(
     server_url: str | None,
     start_page_id: int,
     end_page_id: int | None,
+    image_analysis: bool = True,
 ) -> dict[str, str | list[str]]:
     return _api_client.build_parse_request_form_data(
         lang_list=[language],
         backend=backend,
         parse_method=parse_method,
         formula_enable=formula_enable,
         table_enable=table_enable,
+        image_analysis=image_analysis,
         server_url=server_url,
         start_page_id=start_page_id,
         end_page_id=end_page_id,
@@ -98,6 +100,7 @@ async def run_demo(
     language: str = "ch",
     formula_enable: bool = True,
     table_enable: bool = True,
+    image_analysis: bool = True,
     server_url: str | None = None,
     start_page_id: int = 0,
     end_page_id: int | None = None,
@@ -117,6 +120,7 @@ async def run_demo(
         parse_method=parse_method,
         formula_enable=formula_enable,
         table_enable=table_enable,
+        image_analysis=image_analysis,
         server_url=server_url,
         start_page_id=start_page_id,
         end_page_id=end_page_id,
@@ -225,6 +229,8 @@ def main() -> None:
     formula_enable = True
     # Enable table parsing in the output.
     table_enable = True
+    # Enable image/chart analysis for VLM and hybrid backends.
+    image_analysis = True
     # Required only for "*-http-client" backends, for example:
     # "http://127.0.0.1:30000"
     server_url = None
@@ -245,6 +251,7 @@ def main() -> None:
             language=language,
             formula_enable=formula_enable,
             table_enable=table_enable,
+            image_analysis=image_analysis,
             server_url=server_url,
             start_page_id=start_page_id,
             end_page_id=end_page_id,

demo/office_docs/docx_01.docx

@@ -429,6 +429,12 @@ Text levels are distinguished through the `text_level` field:
 | `list` / `index` | List and index blocks with `list_items` |
 | `page_header` / `page_footer` / `page_number` / `page_aside_text` / `page_footnote` | Page auxiliary blocks |
 
+Inline fields such as `title_content`, `paragraph_content`, and captions are
+usually span lists. A `hyperlink` span contains `content` and `url`; when one
+link contains text fragments with different styles, it may also contain
+`children`. In that case, `content` is the concatenated child text, and the
+exact styles are represented by the child `text` spans.
+
 ##### Sample Data
 
 ```json

demo/office_docs/pptx_01.pptx

@@ -429,6 +429,11 @@
 | `list` / `index` | 列表与索引，包含 `list_items` |
 | `page_header` / `page_footer` / `page_number` / `page_aside_text` / `page_footnote` | 页面辅助块 |
 
+`title_content`、`paragraph_content`、说明文字等行内内容通常由 span 列表组成。
+`hyperlink` span 包含 `content`、`url`，当同一个链接内存在多段不同样式文本时，
+还会包含 `children`；此时 `content` 是 children 文本的拼接，精确样式以
+`children` 中的 `text` span 为准。
+
 ##### 示例数据
 
 ```json

docs/en/reference/output_files.md

@@ -156,6 +156,10 @@ def __init__(
                 block_type = BlockType.INTERLINE_EQUATION
                 span_type = ContentType.INTERLINE_EQUATION
 
+            if span_type == ContentType.TEXT and block_content is None:
+                # 文本类块缺失 content 时按空文本处理，避免 VLM mkcontent 渲染阶段遇到 None。
+                block_content = ""
+
             # code 和 algorithm 类型的块，如果内容中包含行内公式，则需要将块类型切换为 algorithm
             switch_code_to_algorithm = False
 

docs/zh/reference/output_files.md

@@ -0,0 +1 @@
+# Copyright (c) Opendatalab. All rights reserved.