1
00:00:00,000 --> 00:00:11,400
MinerU Popo是一個面向複雜文檔解析的後處理模型框架。核心目標是把OCR和版面分析得到的頁面集結果,進一步整理成文檔集結構化知識。

2
00:00:11,400 --> 00:00:25,800
它並非替代MinerU、Manty OCR或PaddleOCR等項目,而是在這些解析工具進行OCR之後,對標題層級,段落連續性、跨頁表格、圖文關係和閱讀順序進行統一推理與修復。

3
00:00:25,800 --> 00:00:32,600
通俗地說,OCR能力是負責把字看出來。MinerU Popo負責把整份文檔真正組織起來。

4
00:00:32,600 --> 00:00:42,400
在技術上,MinerU Popo採用4B參數後處理模型,並集合任務化數據引擎。動態分塊同步和文檔增強方法。

5
00:00:42,400 --> 00:00:48,400
重點解決長文檔在分翼、跨翼和複雜版式下容易出現的結構斷裂問題。

6
00:00:48,400 --> 00:00:55,400
在公開指標中,接入Popo框架後,MinerU的TEDS從53.7提升到90.6。

7
00:00:55,400 --> 00:01:02,400
MineQ-C2從48.9提升到87.4,說明它能夠顯著提升文檔結構恢復質量。

8
00:01:02,400 --> 00:01:12,400
它的價值在於讓知識庫、RAG智能問答、學術論文解析和企業文檔治理,獲得更乾淨、更穩定、更可計算的文檔資產。

9
00:01:12,400 --> 00:01:18,400
MinerU Popo讓複雜文檔從可識別,走向可理解、可治理、可計算。

10
00:01:18,400 --> 00:01:20,400
MinerU Popo

11
00:01:20,400 --> 00:01:21,400
MinerU Popo

12
00:01:21,400 --> 00:01:22,400
MinerU Popo

13
00:01:22,400 --> 00:01:23,400
MinerU Popo

14
00:01:23,400 --> 00:01:24,400
MinerU Popo

15
00:01:24,400 --> 00:01:25,400
MinerU Popo

16
00:01:25,400 --> 00:01:26,400
MinerU Popo

17
00:01:26,400 --> 00:01:27,400
MinerU Popo

18
00:01:27,400 --> 00:01:28,400
MinerU Popo

19
00:01:28,400 --> 00:01:29,400
MinerU Popo

20
00:01:29,400 --> 00:01:31,400
MinerU Popo

21
00:01:31,400 --> 00:01:32,400
MinerU Popo

22
00:01:32,400 --> 00:01:33,400
MinerU Popo

23
00:01:33,400 --> 00:01:34,400
MinerU Popo

24
00:01:34,400 --> 00:01:35,400
MinerU Popo

25
00:01:35,400 --> 00:01:36,400
MinerU Popo

26
00:01:36,400 --> 00:01:37,400
MinerU Popo

27
00:01:37,400 --> 00:01:38,400
MinerU Popo

28
00:01:38,400 --> 00:01:39,400
MinerU Popo

29
00:01:39,400 --> 00:01:40,400
MinerU Popo

30
00:01:40,400 --> 00:01:41,400
MinerU Popo

31
00:01:41,400 --> 00:01:42,400
MinerU Popo

32
00:01:42,400 --> 00:01:43,400
MinerU Popo

33
00:01:43,400 --> 00:01:44,400
MinerU Popo

34
00:01:44,400 --> 00:01:45,400
MinerU Popo

35
00:01:45,400 --> 00:01:46,400
MinerU Popo