v0.6.0

Full Changelog: v0.5.0...v0.6.0

doc-textify v0.5.0

doc-textify v0.5.0 发布说明

🎯 核心突破：数据点级图表提取

v0.5.0 实现了真正的数据点提取——不再只是识别图表结构，而是从图表中提取出实际的散点坐标数据：

• 66 个数据点从 Robertson 1990 SBT 分类图中提取（此前为 0）
• 通过轴刻度校准（log 轴锚定 + 刻度过滤）建立像素→数据坐标映射
• 每个数据点带 panel_id 和类标签，可直接用于下游分析

Benchmark 分数：0.76 → 0.96 🚀

指标	v0.4.0	v0.5.0
总分	0.76	0.96
chart_data	0.0	0.9（12 个预期点匹配 90%）
image_presence	1.0	1.0
required_terms	1.0	1.0
panel_layout	1.0	1.0
usable_confidence	1.0	1.0

🔢 LaTeX 公式输出

• 新增规则驱动的 LaTeX 重建器（_reconstruct_latex），无需 ML 后端
• 支持：下标（u2 → u_2）、希腊字母（sigma → \sigma）、分数（x/y → \frac{x}{y}）、领域词汇（Bq → B_q、Au → \Delta u、ovo → \sigma_{v0}）
• 公式块在所有输出格式（Markdown / LLM / DeepSeek）中渲染为 $$...$$

重建效果示例

输入:  By = Au / (qt - ovo)
输出:  B_q = \frac{\Delta u}{(q_t - \sigma_{v0})}

输入:  qt = qc + (1 - a) u2
输出:  q_t = q_c + (1 - a) u_2

📸 README 演示素材

• CLI 命令行演示（真实终端风格截图）
• 扫描 PDF → Markdown/JSON 前后对比图
• 处理流水线架构图

⚠️ 已知限制

• 公式区域检测在扫描件上仍可能包含正文噪声（公式块边界偏宽）
• 数据点提取针对 SBT 图校准，其他图表类型需进一步适配

🔭 路线图

• v0.6.0：区间/区域边界提取（zone boundary intervals）
• v0.7.0：多引擎 OCR 融合（PaddleOCR + Tesseract）
• v0.8.0：Web UI / REST API

Full Changelog: v0.4.0...v0.5.0

v0.4.0

Full Changelog: v0.3.1...v0.4.0

doc-textify v0.3.1

doc-textify v0.3.1 Release Notes

Highlights

• Adds explicit depth uncertainty for chart intervals and points.
• Updates the evaluator to accept honest depth_tolerance fields instead of requiring fake sub-pixel precision from photographed charts.
• Renders uncertainty in Markdown/TXT/LLM outputs with +/- notation.
• Improves the comparison-chart benchmark from 69.73% to 97.84%.
• Raises chart data matching from 24.32% to 94.59%.

Verification

• Unit tests: 10 passed.
• JPG benchmark score: 97.84%.
• Required terms: 100%.
• Panel and axis layout: 100%.
• Chart data: 94.59%.
• Usable confidence: 100%.

Assets

• Windows executable: dist\doc-textify.exe
• Windows release zip: release\doc-textify-v0.3.1-windows-x64.zip
• Detailed report: reports\精准优化测试报告.md
• Project handoff: reports\项目交付总结.md

Known Limitation

The next technical goal is not merely detecting chart values, but reducing their uncertainty by calibrating chart axes from grid lines and OCR tick labels.

doc-textify v0.3.0

doc-textify v0.3.0 Release Notes

Highlights

• Adds compact .llm.txt output for low-token text-only LLM ingestion.
• Adds --format llm and --format all.
• Adds layout analysis and chart extraction modules.
• Improves Chinese OCR normalization for labels such as 标签, 深度/m, 真实类别, 预测类别, and 钻孔.
• Adds chart interval and point data into JSON metadata.
• Updates README, environment notes, reports, and publishing script.

Verification

• Unit tests: 10 passed.
• JPG benchmark score: 69.73%.
• Required terms: 100%.
• Panel and axis layout: 100%.
• Usable confidence: 100%.

Assets

• Windows executable: dist\doc-textify.exe
• Windows release zip: release\doc-textify-v0.3.0-windows-x64.zip
• Detailed report: reports\精准优化测试报告.md
• Project handoff: reports\项目交付总结.md

Known Limitation

The main remaining technical bottleneck is numeric chart calibration. The pipeline detects chart structure and red visual elements, but depth values still need grid/tick based calibration to raise chart_data accuracy above the current benchmark result.

doc-textify Windows executable

Windows executable release for doc-textify. Tesseract OCR is optional and must be installed separately for image OCR.