/ppt-analysis

name: ppt-analysis
description: "PPT (.pptx/.ppt) 全量解析。覆盖：所有 slide 文本/表格/图表提取、嵌入图片 caption、纯图片 slide 渲染识别、数据标签提取。"

PPT Analysis — .pptx / .ppt

Environment

from pptx import Presentation
from pptx.util import Inches
import os, subprocess, json

# python-pptx is available
# For .ppt (old binary format): convert via libreoffice
def load_pptx(path):
    if path.lower().endswith('.ppt'):
        import subprocess
        out_dir = os.path.dirname(path)
        subprocess.run(
            ['libreoffice', '--headless', '--convert-to', 'pptx', '--outdir', out_dir, path],
            check=True, capture_output=True
        )
        path = path.rsplit('.', 1)[0] + '.pptx'
    return Presentation(path), path

---

Core Method 1: Full Text Extraction (ALL slides)

def extract_all_slides_text(pptx_path):
    """
    Extract text from every slide: text frames, tables, chart titles.
    For slides with no extractable text, flag them for image captioning.
    """
    prs, _ = load_pptx(pptx_path)
    slides_data = []

    for slide_num, slide in enumerate(prs.slides, start=1):
        slide_texts = []
        has_text = False

        for shape in slide.shapes:
            # Text frame (most common)
            if shape.has_text_frame:
                for para in shape.text_frame.paragraphs:
                    text = para.text.strip()
                    if text:
                        slide_texts.append(text)
                        has_text = True

            # Table
            if shape.has_table:
                tbl = shape.table
                for row in tbl.rows:
                    row_text = '\t'.join(cell.text.strip() for cell in row.cells)
                    if row_text.strip():
                        slide_texts.append(row_text)
                        has_text = True

            # Chart title
            if shape.shape_type == 3:  # MSO_SHAPE_TYPE.CHART
                try:
                    if shape.chart.has_title:
                        title = shape.chart.chart_title.text_frame.text
                        slide_texts.append(f"[Chart: {title}]")
                        has_text = True
                except Exception:
                    pass

        slides_data.append({
            'slide': slide_num,
            'text': '\n'.join(slide_texts),
            'has_text': has_text,
            'needs_caption': not has_text  # flag image-only slides
        })

    print(f"Total slides: {len(slides_data)}")
    image_only = sum(1 for s in slides_data if s['needs_caption'])
    print(f"Slides with text: {len(slides_data) - image_only}, image-only: {image_only}")
    return slides_data

---

Core Method 2: Table Extraction (Structured)

import pandas as pd

def extract_pptx_tables(pptx_path):
    """Extract all tables from all slides as DataFrames."""
    prs, _ = load_pptx(pptx_path)
    all_tables = []

    for slide_num, slide in enumerate(prs.slides, start=1):
        for shape in slide.shapes:
            if not shape.has_table:
                continue
            tbl = shape.table
            rows = []
            for row in tbl.rows:
                rows.append([cell.text.strip() for cell in row.cells])

            if not rows:
                continue

            # Use first row as header
            try:
                df = pd.DataFrame(rows[1:], columns=rows[0])
            except Exception:
                df = pd.DataFrame(rows)

            all_tables.append({'slide': slide_num, 'df': df})
            print(f"  Slide {slide_num}: table {df.shape[0]}r × {df.shape[1]}c")
            print(df.head(3).to_string())

    return all_tables

---

Core Method 3: Chart Data Extraction

`python-pptx` can read Chart data when it's stored as embedded Excel data. If that fails, fall back to captioning the slide image.

def extract_chart_data(pptx_path):
    """
    Extract data series from Chart shapes.
    Returns list of {slide, chart_title, series_name, categories, values}.
    """
    prs, _ = load_pptx(pptx_path)
    charts = []

    for slide_num, slide in enumerate(prs.slides, start=1):
        for shape in slide.shapes:
            if shape.shape_type != 3:  # not a chart
                continue
            try:
                chart = shape.chart
                title = chart.chart_title.text_frame.text if chart.has_title else f"Chart_S{slide_num}"

                for plot in chart.plots:
                    for series in plot.series:
                        try:
                            categories = [str(pt.label) for pt in series.data_labels] if hasattr(series, 'data_labels') else []
                            values = [pt.value for pt in series.values] if hasattr(series, 'values') else []
                            # Alternative: use xChart data
                            if not values:
                                values = list(series.values)
                        except Exception as e:
                            values = []
                            categories = []

                        charts.append({
                            'slide': slide_num,
                            'chart_title': title,
                            'series': getattr(series, 'name', ''),
                            'categories': categories,
                            'values': values
                        })
            except Exception as e:
                print(f"  Slide {slide_num}: chart extraction failed ({e}) — will use caption")

    return charts

---

Core Method 4: Render Image-Only Slides → Caption

When a slide has no extractable text (pure image/screenshot slides):

import fitz  # PyMuPDF can also render PPTX via LibreOffice conversion

CAPTION = "/path/to/skills/sn-da-image-caption/scripts/caption.py"

def caption_image_slides(pptx_path, slides_data, prompt=None):
    """
    For slides flagged as 'needs_caption', render to PNG and caption.
    Uses LibreOffice